IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO.

3, SEPTEMBER 2008 539

Digital Single Lens Reflex Camera Identification

From Traces of Sensor Dust
Ahmet Emir Dirik, Husrev Taha Sencar, and Nasir Memon

Abstract—Digital single lens reflex cameras suffer from a well- A key problem in media forensics is the identification and
known sensor dust problem due to interchangeable lenses that they analysis of media characteristics that relate to the acquisition
deploy. The dust particles that settle in front of the imaging sensor device. These characteristics are essentially a combination of
create a persistent pattern in all captured images. In this paper, we
propose a novel source camera identification method based on de- two interrelated factors: 1) the class properties that are common
tection and matching of these dust-spot characteristics. Dust spots among all devices of a brand and model and 2) the individual
in the image are detected based on a (Gaussian) intensity loss model properties that set a device apart from another in its class. Hence,
and shape properties. To prevent false detections, lens parameter- research efforts have focused on the design of techniques to
dependent characteristics of dust spots are also taken into con- identify class and individual characteristics of data-acquisition
sideration. Experimental results show that the proposed detection
scheme can be used in identification of the source digital single lens devices without requiring a specific configuration of source de-
reflex camera at low false positive rates, even under heavy compres- vices [6], [7].
sion and downsampling. Two principal research approaches have emerged in the
Index Terms—Digital forensics, digital single lens reflex (DSLR), effort to establish characteristics that can link an image or video
sensor dust. to its source. The first approach focuses on determining the
differences in processing techniques and component technolo-
gies. For example, optical distortions due to a type of lens, the
I. INTRODUCTION size of the imaging sensor, the choice of color filter array and
corresponding demosaicing algorithm, and color-processing
algorithms can be detected and quantitatively characterized by

G IVEN the fast and widespread penetration of multimedia

into all areas of life, the need for mechanisms to en-
sure reliability of multimedia information has become impor-
appropriate image-analysis techniques [3], [8]–[12]. The main
difficulty with this approach is that many device models and
brands use components by a few manufacturers and processing
tant. Today, digital media is relied upon as the primary way to methods remain the same, or very similar, among different
present news, sports, entertainment, and information regularly models of a brand. Hence, reliable identification of class char-
that captures current events as they occur. They are introduced acteristics of a device requires consideration of many different
as evidence in court proceedings and commonly used in pro- factors.
cessing, analysis, and archiving of financial and medical doc- In the second approach, the primary goal is to identify unique
uments. The long-term viability of these benefits requires the characteristics of the source-acquisition device. These may be
ability to provide certain guarantees about the origin, veracity, in the form of hardware and component imperfections, defects,
and nature of the digital media. For instance, the ability to estab- or faults which might arise due to inhomogeneity in the manu-
lish a link between a camera and the digital image is invaluable facturing process, manufacturing tolerances, environmental ef-
in deciding the authenticity and admissibility of a digital image fects, and operating conditions. The ability to reliably extract
as legal evidence. Similarly, doctoring images is becoming more these characteristics makes it possible to match an image or
frequent as a way to influence people and alter their attitudes video to its potential source and cluster data from the same
in response to various events [1], [2]. Hence, for conventional source device together. The main challenge in this research di-
and online media outlets, the capability to detect doctored im- rection is that reliable measurement of these minute differences
ages before they are published is important to maintain cred- from a single image is difficult and can be easily eclipsed by
ibility. Recent research efforts in the field of media forensics the image content itself. Another challenge is that these arti-
have begun to address these issues [3]–[5]. facts tend to vary in time and depend on operating conditions;
therefore, they may not always yield positive identification. To
Manuscript received June 30, 2008; revised April 15, 2008. First published date, proposed methods in this area depend primarily on faulty
July 9, 2008; last published August 13, 2008 (projected). This work was sup-
ported by the National Institute of Justice under Grant 2006-92251-NY-IJ. The
elements of the imaging device [13] and noise characteristics of
associate editor coordinating the review of this manuscript and approving it for the imaging sensor [12], [14]–[18].
publication was Dr. Jessica J. Fridrich. In this paper, we present a new approach to source camera
A. E. Dirik is with the Department of Electrical and Computer En-
gineering, Polytechnic University, Brooklyn, NY 11201 USA (e-mail:
identification considering digital single-lens reflex (DSLR)
[email protected]). cameras. The basis of our method is the appearance of dust
H. T. Sencar and N. Memon are with the Information Systems and Internet spots or blemishes in DSLR camera images. Based on our
Security Laboratory, Polytechnic University, Brooklyn, NY 11201 USA. earlier work [19], we demonstrate how these artifacts can be
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. utilized as a fingerprint of the camera. DSLR cameras differ
Digital Object Identifier 10.1109/TIFS.2008.926987 from digital compact cameras in various aspects: larger and
1556-6013/$25.00 © 2008 IEEE
540 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO. 3, SEPTEMBER 2008

Fig. 1. Sensor dust appears in two different images taken with the same DSLR camera. Local histogram adjustment is performed to make dust spots visible (2nd
row). White boxes show dust-spot positions.

photo sharing websites. For instance, the top five cameras for
November 2007 in Flickr (flickr.com) and Pbase (pbase.com)
photo sharing websites are all DSLR cameras.
The very nature of a DSLR camera allows users to work with
multiple lenses, but this desirable feature creates a unique and
undesired problem. Essentially, during the process of mounting/
unmounting the interchangeable lens, the inner body and work-
ings of the camera is exposed to the outside environment. When
the lens is detached, very small dust particles are attracted to the
camera and settle on the protective element (dichroic mirror or
low-pass filter) in front of the sensor surface. These tiny specks
of dust, lint, or hair cling to the surface and form a dust pat-
tern which later reveals itself on captured images as blemishes
or blotches. We will refer to this type of artifact as dust spots in
the rest of this paper. Dust spots on two different images.2 taken
with same DSLR are given in Fig. 1. To make dust spots more
visible, each pixel color is changed through histogram equaliza-
tion in windows of small size. Dust spots become visible at low
Fig. 2. Dust spots may stay in the same position for years. aperture rates (e.g., for high -numbers) since a large aperture
will allow light to wrap around the dust particles and make them
out of focus. Moreover, sensor dust is persistent and accumula-
higher quality sensors with low noise power, parallax-free tive and unless it is cleaned, it may remain in the same position
optical viewfinder that allows error-free viewing of the scenery, for a very long time, as exemplified by images3 in Fig. 2.
less shutter lag, interchangeable lenses, and a better control To deal with the sensor dust problem, various solutions
over depth of field. According to the 2006 International Data have been proposed. Some DSLR camera manufacturers have
Corporation (IDC) report on the digital camera market, DSLR already incorporated built-in mechanisms for dust removal. For
cameras showed a consistent growth with a total market and a example, Sony’s Alpha A10 DSLR uses an antidust coating on
39% increase from the 2005 figure.1 Not surprisingly, DSLR
2[Online]. Available: www.pbase.com/chucklantz/image/38843803 and
cameras also take the top place in most popular camera lists of
\ldots /72922162.
1[Online]. Available: www.imaging-resource.com/NEWS/1175724860. 3[Online]. Available: www.pbase.com/chucklantz/image/47472873 and
html. \ldots /image/72922162.
DIRIK et al.: DIGITAL SINGLE LENS REFLEX CAMERA IDENTIFICATION FROM TRACES OF SENSOR DUST 541

the CCD with a vibrating mechanism which removes the dust Section IV. The robustness of the proposed scheme to compres-
by shaking it. Similar vibration mechanisms are also utilized in sion and downsampling is explained in Section IV. Finally, our
Olympus E-300 and Canon EOS Rebel DSLR cameras. Nikon conclusions are presented in Section V.
D50 and Canon Digital Rebel also offers a software solution to
remove dust spots by creating a dust template of the camera. A. Related Work
A comprehensive benchmark on the performance of built-in The first work in the field of source identification was un-
dust removal mechanisms has been performed by pixinfo.com.4 dertaken by Kurosawa et al. [14] for camcorders. Their method
The study involved four of the state-of-the-art DSLR cameras, relies on the fact that each digital camcorder CCD sensor has
namely, Canon EOS-400D, Olympus E-300, Pentax K10D, and a unique and intrinsic dark current noise pattern. This specific
Sony Alpha DSLR-A10. In the experiments, these four cameras noise pattern reveals itself in the form of fixed offset values in
were initially exposed to the same dusty environment, and later pixel readings, and it can be easily extracted when the sensor
the cameras’ built-in functions were used to remove these dust is not exposed to any light. However, the drawback of this ap-
particles. Their results show that even after 25 consecutive invo- proach is that today, cameras are designed to compensate for
cation of the cleaning mechanism, dust spots were still present this type of an artifact. Later, Geradts et al. [13] proposed using
and their performance was far from satisfactory.5 sensor imperfections in the form of hot and dead pixels, pixel
Although vibration-based internal cleaning mechanisms do traps, and pixel defects in order to match images with cam-
not work satisfactorily, they might influence the positions of eras. Although their results show that these imperfections are
dust particles over the filter component. This phenomenon can unique to imaging sensors and they are quite robust to JPEG
also be observable from the benchmarks mentioned before. To compression, most digital cameras, today deploy mechanisms
quantify the effect of internal cleaning mechanisms on dust-spot to detect and compensate pixel imperfections through postpro-
positions, the proposed dust detection algorithm was applied to cessing, which restricts the applicability of their technique.
two blank images taken with Canon EOS-400D after 2nd and Recently, similar to [14], Lukáš et al. [15] and Chen et al.
25th cleaning operations. These two images were obtained from [16], [17] proposed a more reliable sensor noise-based source
the cleaning benchmark experiments in pixinfo.com. Once dust digital camera and camcorder identification method. Their
positions are detected, they were compared with each other. method is based on the extraction of the unique photoresponse
After the 25th cleaning, %97.01 (24 dust particles) out of 803 nonuniformity (PRNU) noise pattern which is caused by the
dust particles remain in the same position. The maximum de- impurities in silicon wafers and sensor imperfections. These
tected position shift after the 25th cleaning is 5.83 pixels (image imperfections affect the light sensitivity of each individual
size is ). Since dust shifts due to internal cleaning pixel and cause a fixed noise pattern. Similarly, Khanna et al.
mechanisms are not significant, we will omit the effect of filter [18], Gou et al. [12], and recently Gloe et al. [20] have ex-
vibrations on dust positions in the rest of this paper. tended PRNU noise extraction methodology to source scanner
An alternative solution is the manual cleaning of the dust by identification where the imaging sensor is typically a 1-D linear
using chemicals, brushes, air blowing, and dust adhesive. Al- array. The drawback of this approach is that it is very hard to
though these are known to be more effective, manual cleaning synchronize the scanner noise pattern with the noise residue
is a tedious task and may potentially harm the imaging sensor; extracted from the scanned image. This is due to difficulty in
therefore, it is not recommended by camera manufacturers.6 controlling the document position during scanning. Therefore,
In this paper, we exploit this persistent nature of the sensor authors extracted statistical characteristics of PRNU noise and
dust to match DSLR images to their sources. The matching can deployed machine learning methods to identify the scanner
be realized by obtaining a dust pattern directly from the camera brand and model. It should be noted that utilizing feature-based
or from a number of images taken by the camera, as in Fig. 1. classifiers makes these methods less effective in individual
It should be noted that since the sensor dust problem is solely source scanner identification.
intrinsic to DSLR cameras, the detection of any sensor dust in
a given image can be taken as strong evidence that the image II. SENSOR DUST CHARACTERISTICS
source is a DSLR camera. In addition, by detecting traces of Essentially, dust spots are the shadows of the dust particles
sensor dust, it may be possible to order images, taken at different in front of the imaging sensor. The shape and darkness of the
times, in capturing time by evaluating accumulation character- dust spots are determined primarily by the following factors:
istics of dust. distance between the dust particle and imaging sensor, camera
The rest of this paper is organized as follows. In Section II, we focal length, and size of aperture. A general optical model
investigate the optical characteristics of sensor dust as a function showing the formation of dust spots is given in Fig. 3. When
of imaging parameters. In Section III, a model-based dust-spot the focal plane is illuminated uniformly, all imaging sensors
detection method and its use in source camera identification will yield the same intensity values. However, in the presence
is explained in detail. The efficacy of the proposed method is of sensor dust, light beams interact with the dust particles and
substantiated by experimental results for two different cases in some of the light energy is absorbed by the dust particles.
4[Online].
The amount of the absorbed energy is directly related to the
Available: pixinfo.com/en/articles/ccd-dust-removal/.
5The
parameter -number (F/#) which is defined as the ratio between
reported dust removal performances defined based on a successfully
cleaned number of initially present dust spots are as follows: Olympus E-300: the focal length and the aperture
50%, Canon EOS-400D: 5%, Pentax K10D: 0%; and Sony Alpha A10: 0%.
6[Online]. Available: www.usa.canon.com/consumer. -number (1)
542 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO. 3, SEPTEMBER 2008

Fig. 3. Intensity degradation due to the dust spot.

At small apertures and high -numbers, the light source can

be assumed to be a pinpoint source resulting in a relatively
narrow light cone which can be blocked mostly with a tiny
sensor dust. As a result, a strong dust shadow will appear on the
image. This phenomenon is illustrated in Fig. 3(a). On the other Fig. 4. Spot of hair/lint for different f -numbers (Nikon D50).
hand, for wide apertures or small -numbers which cause wide
light cones in the DSLR body, most light beams pass around
the dust spots causing a blurry and soft blemish in the image. with unique shapes, are likely to cause very large intensity
In Fig. 3(b) and (c) and Fig. 6, the actual intensity degradations degradations, they are easily noticeable. Although this type of
caused by dust spots are shown for different -numbers. It can sensor dust is very suitable for camera identification, it is likely
be seen from the figures that the change in -number affects the to attract the user’s attention due to their annoying appearance.
intensity and radius of the dust spot wherein an increase in the As a result, they are more likely to be cleaned out. Therefore,
-number (smaller aperture) causes dust spots to appear darker in this paper, we will focus on dust spots due to much smaller
and smaller. particles that yield round-shaped dust spots that are less likely
to be cleaned by many users and are, in fact, difficult to clean
A. Dust-Spot Shape as discussed earlier.
The dust particles in front of the imaging sensor mostly ap-
pear as round-shaped blemishes (see Figs. 1 and 2). However, B. Dust-Spot Size
dust spots with different shapes are also possible due to larger In this section, the formation of dust spots as a function of
particles, such as lint or hair (Fig. 4). Since these large spots, the camera parameters is analyzed. The optical dust-spot model,
DIRIK et al.: DIGITAL SINGLE LENS REFLEX CAMERA IDENTIFICATION FROM TRACES OF SENSOR DUST 543

TABLE I
DUST-SPOT PROPERTIES FOR DIFFERENT -f NUMBERS (f = 55 mm,
2
NIKON D50). IMAGE DIMENSIONS:1504 1000

which denotes the center of the dust shadow. To see how the
dust shadow center is related to camera parameters, and
, which are computed in (2) and (3), are substituted into the
formula and is obtained as

(5)

Fig. 5. Optical dust-spot model. (6)

where and define the aperture size. The equation shown

before implies that the dust-spot position is not affected by the
aperture but rather by the focal length. Indeed in Fig. 6, blemish
center positions do not change with different apertures. How-
ever, the focal length can be changed with a zoom lens and dif-
Fig. 6. Same dust spot for different f -numbers. The focal length is fixed to
55 mm (Nikon D50). ferent focal lengths may shift the dust spots.
Let us define the dust-spot shift as the distance between
dust-spot centers , where and are the
we assume, is depicted in Fig. 5, where the parameters , , , dust-spot centers with focal length and , respectively. By
, and refer to the aperture, focal length, filter width, dust substituting (6) into the definition, the dust-spot shift is obtained
diameter, and dust shadow diameter, respectively. Assuming a as
circular dust spot, its size on the image can be computed. Let
and points define the diameter of the dust shadow on the (7)
imaging sensor; and let and points define the diameter of
the actual dust particle . From the similarity of triangles, (8)
and can be written in terms of focal length, aperture, and the
distance of dust particle ( and ) to the image optical center It is seen from the equation that the dust-spot shift is directly
(see Fig. 5) as follows: proportional to the radial dust position and the filter width .
However, in (8) the relation between the focal length change
and the dust-spot shift is not clear. To visualize this
(2) relation, (8) was evaluated over a set of different focal length
and values for a fixed (55 mm). The evaluation results are
(3) depicted in Fig. 8.
It can be seen from Fig. 8 that the shift in a dust spot depends
Hence, the dust-spot (shadow) diameter on the on the focal length change reciprocally. Besides, the
imaging sensor becomes dust-spot shift magnitude is also determined by the actual dust
position on the filter component. The farther the dust is
(4) from the image origin , the higher the shift with the change in
focal length is. Apparently, the shift vectors lie along the image
Essentially, (4) states that the size of the dust spot is directly pro- radial axes. To measure the shift vectors from real images, we
portional to the DSLR aperture , which agrees with the obser- assume that the origin of the image is also the optical center
vation in Fig. 6. Similarly, Table I and Fig. 7 show the change in of the image. In Fig. 9, the dust shift phenomenon is illustrated
diameter of a dust spot for fixed focal length and different aper- where a reduction in focal length causes dust spots in the image
tures. It can also be seen from the table that the dust-spot size to move outward along the radial axis.
decreases with a decrease in aperture. Experimental results for measuring dust-spot shifts are given
in Table II and Fig. 10 for a Nikon D50 DSLR camera. In mea-
C. Dust-Spot Movement suring the shifts, two different images (with 1504 1000 reso-
Although actual dust positions are stable on an imaging lution) were taken at two different focal lengths of 18 mm (F/18)
sensor, the positions of dust spots are affected by -number and 55 mm (F/36). From these images, four distinct dust spots
changes. To see how a dust-spot position changes with aperture were determined. Their radial positions from the image center
and focal length , a new variable is defined were measured in polar coordinates. The radial shift magnitudes
544 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO. 3, SEPTEMBER 2008

Fig. 7. Dust-spot properties for different f -numbers.

(a) (b)

Fig. 8. Dust-spot movement analysis based on the proposed optical model. (a) Radial shifts of two dust spots. t and f are fixed. f is changed. (b) Dust-spot
shifts for different t values.

TABLE II
DUST-SPOT POSITIONS AND SHIFTS FOR DIFFERENT
FOCAL LENGTHS (NIKON D50)

and angles are given in the last columns for four dif-
ferent dust spots. It is seen from the table that the results are
consistent with (8). From (8), it is possible to estimate the shift
magnitudes. Since the parameter of Nikon D50 is not known,
first, was estimated from the observed dust shifts in Table II as
0.35 mm by the least square method. Then, for each dust spot
in the table, the shift magnitudes were estimated from (8) with
Fig. 9. Dust-spot shifts due to focal length f change.
1.2 pixel mean absolute error. The estimation results are given
with the actual values in Fig. 10.
DIRIK et al.: DIGITAL SINGLE LENS REFLEX CAMERA IDENTIFICATION FROM TRACES OF SENSOR DUST 545

intensity surface as a function of the -number and 2) they

appear mostly in the form of rounded shapes.
As mentioned before, sensor dust can be viewed as black,
out-out-focus spots with a soft intensity transition. Our obser-
vations of various actual dust spots also confirm that they have
Gaussian-like intensity degradations. This phenomenon can be
viewed in Fig. 3(b) and (c). Inspired from these figures and many
other examples, we utilize a Gaussian intensity loss model (i.e.,
a 2-D Gaussian function to model dust spots). Our model for the
dust spot is expressed as follows:

intensity loss
(9)

where , , and are the gain factor, standard deviation, and

template width, respectively.
Essentially, dust-spot dimensions depend on the -number
and dust size directly [see (4)]. To capture this relation in our
Fig. 10. Estimated and observed dust-spot shifts for different focal lengths and model, is selected to adjust the size of the dust-spot model.
positions.
The intensity loss of the model is controlled by the parameter
. Although, for a given image, the -number can be obtained
from the EXIF data, the actual dust size cannot be known. Nev-
III. FORENSICS USE OF SENSOR DUST ertheless, to investigate the more general case, we do not utilize
In this section, we develop a technique for camera identifica- the EXIF header information for this purpose. Hence, the model
tion based on sensor dust detection. The use of dust spots for parameter that determines the dust-spot size needs to be esti-
source camera identification first requires determining the posi- mated blindly.
tions of dust spots in an image. Since dust particles do not tend to In detecting dust spots in an image, we correlate the Gaussian
move easily, they appear in all images taken with high -num- dust model with the image for various values (9) over all pixel
bers, and their proper cleaning is not trivial, these dust-spot lo- positions via fast normalized cross-correlation (NCC) [25]. This
cations can be used as a unique fingerprint of a DSLR camera. results in a 2-D map of values where each value is computed
This fingerprint can be represented by a camera dust template by cross-correlating the Gaussian dust model with a window of
that includes information on all detectable dust spots. It must size sliding over the image, which will be referred to
be noted that this template can be directly obtained from im- as NCC output.
ages taken by the camera at a high -number setting or from a In the NCC output, values higher than an empirically set
number of images when the camera is not available by collating threshold are selected as potential dust-spot candidates. To re-
dust spots detected from different images together. To decide duce the search range of the parameter and to speed up the
whether an image is taken by a given DSLR, the dust spots de- detection process, all images are suitably downsampled to a
tected in an image can be compared to those in the camera dust midresolution (800 533) while preserving their aspect ratio.
template and a decision is made depending on the match be- In addition, to further simplify the processing, images are con-
tween detected dust spots and the template. It should be noted verted to gray level while still preserving intensity degradations
that the lack of a match does not allow a conclusive decision due to dust spots.
since dust specks might have been cleaned manually after the In Table III, the largest dust-spot dimensions observed in
capture of the image with cloning tools. seven different cameras with various -numbers are given.
All dust dimensions were measured after downsampling (to
800 533 pixels resolution). Although the information in
A. Dust-Spot Detection
Table III is not sufficient to represent dust-size distribution, it
In recent years, several software-based dust-spot detection can be used in selecting a range for values.
and removal schemes have been proposed [21]–[24]. These Our measurements in Table III indicate that dust spots gener-
methods usually aim at detecting dust positions from a flat ally have an area that is less than that of a window of 10 10
background by examining intensity gradient degradations. pixels. Based on this observation, we chose or
However, our experimental studies show that gradient-based which correspond to small (6 6 pixels) and large (12 12
approaches suffer from relatively high false detection rates pixels) dust spots, respectively.
due to their sensitivity to intensity variations. Alternatively, To exemplify the relation between the NCC output and model
in this paper, and based on our earlier work [19], we propose parameter , our dust detection scheme was applied to various
a model-based dust detection scheme that utilizes dust-spot dust spots. In Table IV, NCC local maxima values computed
intensity and shape characteristics in detecting dust spots. In through our detection scheme for different dust spots taken
our proposed detection scheme, we model dust spots based on from various DSLR cameras (2 2 pixels to 20 20 pixels)
their two major characteristics: 1) an abrupt change on image are given. As seen from the table, the corresponding NCC local
546 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO. 3, SEPTEMBER 2008

TABLE III 1) Binary Map Analysis: For a given image, NCC values
MAX. DETECTED DUST-SPOT SIZE FOR DIFFERENT
DSLR CAMERAS (IMAGE SIZES: 800 533) 2 are computed for each Gaussian dust model corresponding to
different values. Then, a binary dust template is generated by
thresholding the correlation values such that values smaller than
a preset value are set to zero and others to one. In the binary
dust map, each binary object, obtained by combining together
neighboring binary components, is indexed and a list of dust-
spot candidates is formed. We then exploit the fact that most
dust spots have rounded shapes. This is realized by computing
the area of each binary object and removing the extremely large
TABLE IV or line-shaped objects, resulting in edges and textures from the
MAX. NORM. CROSS-CORRELATION (NCC) OUTPUTS binary dust map.
FOR VARIOUS DUST-SPOT (IMAGE SIZES: 800 533) 2 2) Validation of Correlation Results: After binary map anal-
ysis, all detected dust spots are re-evaluated by analyzing the
values in the NCC output. For actual dust spots, NCC values
are expected to monotonically decrease around the center of the
dust spot (see Figs. 11 and 18). For this, several NCC values
around each binary object are checked along a circular path to
ensure that NCC values exhibit such a decrease. The binary ob-
jects that do not confirm to this observation are also removed
from the binary dust map.
3) Spatial Analysis: The spatial intensity loss characteristics
of each dust-spot candidate (e.g., remaining binary objects in the
binary dust map) is examined by constructing a contour map of
a region surrounding each candidate dust spot and counting the
number of local minima. If there is a global minimum in the se-
lected region, the corresponding binary object is tagged as dust.
On the other hand, the presence of the multiple local minima im-
plies that detected dust-spot candidates are most likely the result
of image content and, therefore, corresponding binary objects
are removed in the final binary dust map.

B. Camera Dust Template Generation

Due to difficulties in dust detection, template generation can
be a challenging task. For instance, differentiating the slight in-
tensity variations due to sensor dust in highly textured regions
Fig. 11. Normalized cross-correlation (NCC) outputs for different dust spots. from the image content is not trivial. Similarly, at large aper-
2 2
4 4 [pixels] dust (left), 13 13 [pixels] dust (right). tures, most dust spots become almost invisible without signifi-
cant intensity degradation. However, in cases where the camera
is available, these problems can be easily circumvented as the
maxima of dust spots takes values between 0.44 and 0.81 which user can adjust the camera settings to make the dust spots visible
are sufficiently high for dust-spot detection. To visualize the as much as possible. (This can be achieved by taking a bright
spatial NCC output variations, NCC mesh plots of two different and unfocused sky photograph with the highest -number set-
dust spots are given in Fig. 11. In the figure, our dust model ting.) This would make almost every dust in the camera appear
produces Gaussian-like NCC outputs with high NCC values as black, tiny spots on a flat background. In such a case, even
at the center of dust spots (0.81 for small dust and 0.53 for a one image is sufficient to create a quite reliable camera dust tem-
relatively large one). In the camera identification phase, the plate.
proposed dust-spot detection scheme is repeated for each On the other hand, if the DSLR camera is not accessible but
value (i.e., ). Then, all detected dust-spot positions are only a set of images taken with the source camera in question
combined together to detect various-sized dust spots in a given are available, the camera template can be estimated by com-
image. bining all detected dust spots from all images in the set. To
Obviously, the described template matching-based method create a camera dust template, the dust detection procedure is
is likely to detect some content-dependent intensity degrada- applied to all available images that are known to be taken with
tions as dust spots. To reduce content-dependent false detec- the same DSLR camera. Since these images may be taken with
tions, template matching is applied only in low and medium de- different f-lengths, detected positions for dust spots may not
tail image regions determined through measurement of intensity overlay with each other, see (8). This misalignment is due to
gradients. Further, the following steps are performed to reduce different -number values. To be able to deal with the shifts
false detections. in dust spots, in creating the binary map, we allow each dust
DIRIK et al.: DIGITAL SINGLE LENS REFLEX CAMERA IDENTIFICATION FROM TRACES OF SENSOR DUST 547

(a) (b) (c)

Fig. 12. Dust template generation from a set of images (Canon EOS Digital Rebel). (a) Image used to create the dust template. (b) Upper left portion of the dust
template. Its actual size is shown at the left. (c) Binary version of the dust template.

candidate position to occupy a circle rather than assigning fixed C. Camera Identification
coordinates.
As can be seen in (8), the dust-spot shift magnitude is di- The final step of DSLR camera identification is done by
rectly proportional to the filter width . This entails that the matching the dust spots detected in an image with dust spots
largest radial shift may vary among DSLR cameras with dif- in the camera dust template. The identification process is
ferent brands/models. Hence, in generating the template, the comprised of three steps:
radius of the binary circle is determined empirically by mea- Step 1) dust-spot detection and matching;
suring the largest radial shifts of dust spots in several images of Step 2) computing a confidence value for each matching
different DSLR cameras. At the end, all binary dust maps are dust spot;
simply added up to create the final camera dust template. Step 3) decision making.
To exemplify camera dust template generation, ten images In the first step, dust spots are detected as explained in
taken with different -numbers were used. The DSLR camera Section III-A. Once dust spots are located, each dust position
used in this experiment was a Canon EOS Digital Rebel. In all is matched with the dust positions in the camera dust template.
images, dust spots were determined and all results were com- The comparison is realized by measuring Euclidian distances.
bined to create the camera dust template. To eliminate the false If the distance is lower than a predetermined value, the corre-
detections in the template, we utilized a threshold. If a dust spot sponding dust position is added to the matching dust-spot list.
appears in only one image and does not appear in other images In the second step, three metrics are computed for each of the
used in template generation, that spot is removed from the dust matching dust spots as follows.
template. 1) The dust occurrence metric is the number of coinciding
The upper left part of the final dust template obtained is shown dust for the corresponding dust spot in the dust template.
in Fig. 12. In Fig. 12(a), the number of coinciding dust spots Higher values of correspond to salient dust spots.
is given. The hot colors refer to high number of dust matches. 2) Smoothness metric presents the smoothness of the re-
In the figure, the dust shifts due to different focal lengths can gion in which a dust spot was detected. Measuring the
be seen clearly. In Fig. 12(b), the binary version of the final amount of local intensity variations is essential in making
dust template is given. Final dust positions were computed as decisions since dust-spot detection in smooth regions is
centroid points of each dust region in the binary map. These more reliable than in busy regions. This is computed via
points are represented with the “+” symbol in Fig. 12(b). the intensity gradient around the dust spot as a binary value.
After template generation, all dust spots in the dust template For a smooth region, becomes one, and for a nonflat or
are tagged with different numbers. Dust centroid positions and nonsmooth region around the dust spot, it becomes zero.
the number of coinciding dust in that positions are saved in a 3) Shift validity metric indicates the validity of a dust spot
file to be used in camera identification. It is assumed that the based on the shift it exhibits. To compute , we do the
higher the number of coinciding dust spots is, the more domi- following.
nant the corresponding dust spot will be. Therefore, those dust • Each dust spot in the matched dust-spot list is tracked
spots will be given more weight in making a decision. In ad- in all template images, used in template generation. (It
dition, all dust positions detected in each individual image are should be noted a different subset of dust spots will be
also maintained since the camera dust template contains only detected in each image.)
averaged dust locations. This information could not be used in • For each dust spot in the list, a set of the shift vectors
detecting the dust-spot shifts properly since we lose individual (i.e., magnitude and angle) is computed by measuring
dust positions for different -numbers after computing centroid the shifts between a dust spot and its matched counter-
positions in a binary dust template [see Fig. 12(b)]. parts in the template images).
548 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO. 3, SEPTEMBER 2008

• Shift vectors associated with each of the template im- TABLE V

ages are collected together. CAMERA DUST TEMPLATE GENERATION
The underlying idea is that since the template images and the
image in question are likely to be taken at different -numbers,
the relative shift between the dust spots in the image and any of
the template images should be consistent. (Fig. 12 displays part
of a camera dust template and its binary version. Since each tem-
TABLE VI
plate image is captured under a different -number, the detected EFFECT OF THE VARIATION IN THE DUST
dust spots appear shifted and, as a result, they do not align in the TEMPLATE TO IDENTIFICATION ACCURACY
template.) In other words, each shift vector along the radial axis
should be in the same direction (i.e., all shifts should be either
towards or outwards from the optical center). (Note that we as-
sume the image center is the optical center.) If a measured shift
vector is significantly away from the radial axis, that dust spot is
ignored and not used for source matching. Similarly, for a given
dust spot, if a significant shift magnitude is not measured, is
IV. EXPERIMENTS
assigned a zero value; otherwise, .
Essentially, the higher the aforementioned three metrics are, To test the efficacy of proposed scheme, several experiments
the more likely that dust detection is correct. If a dust spot is were performed with various DSLR cameras and image sets.
detected in a smooth region, becomes one. For that dust spot, Before starting the experimental analysis, we compute the
if the shift is valid, becomes one too. Finally, if the detected upperbound on false dust detection probability. Let be the
dust spot corresponds to a region where a dominant dust spot lies amount of dust in a camera dust template and all dust spots
where many dust spots coincide, takes as value the number be represented with circles of radius in the template. We
of coinciding dust spots in the template. To bound between assume here that all dust spots are uniformly distributed in the
zero and one, as and , it is applied to a monotonically template. When one pixel position in an image is randomly
increasing function whose upper bound is one. In this work, picked, the probability of that the pixel not coinciding with any
we used Gauss error function as the normalization function. For of dust spots in the template can be computed as
perfect dust detection, the sum of , , and normalized
metrics becomes three for one dust spot. (12)
In the third step, the three metrics computed for each dust
spot are then combined to determine an overall confidence in where and are the image dimensions. Hence, when
identification of the source DSLR. Finally, to make a decision, pixel positions are chosen randomly from the image, the proba-
the statistic is obtained by summing up the confidence score bility of at least one of them coinciding with a dust spot becomes
of all dust spots as
(13)

For , , , , , be-
comes 0.0349. It should be noted by requiring a greater number
of random matches (as opposed at least 1), this figure can be
further reduced. In addition, dust-spot shape characteristics and
(10) shift analysis make it possible to reduce the false detection rate
(11) to lower values.

A. Effect of Image Content on Dust Template

where is the number of detected dust spots in a given image,
Creating an accurate and error-free camera dust template is
is a monotonically increasing normalization function which
very essential for the identification step. If the camera itself is
takes values between 0 and 1, is a scaling factor, and is a
available, the detection of dust spots and dust template genera-
step function defined in (11), respectively. (We select to be the
tion is very straightforward. By adjusting the focal length to its
Gauss error function, but any monotonically increasing bounded
maximum value, almost all dust spots can become visible and
function can be alternatively used.) The reason we used a step
easily detectable. However, in the absence of the camera, dust
function in (10) is to validate the dust shift direction based on
template generation is highly affected by the content of the im-
the availability of at least two dust spots. If two or more dust
ages used in template generation. Intuitively, images with large
spots shift consistently, we infer that dust-spot candidates are
and smooth regions would yield more accurate dust template
not false positives. To make a decision, we apply a threshold
than images with busy content.
to the confidence value. The images which yield the confidence
To determine this, we conducted an experiment with 110 im-
values above the detection threshold are assumed to be taken
ages downloaded from the web.7 From the EXIF image headers,
with the DSLR camera from which the dust template is gener-
it was deduced that the images were taken with a Sigma SD9
ated. On the other hand, low confidence values do not imply that
they are not taken with the suspected DSLR. 7[Online]. Available: www.pbase.com/chucklantz/.
DIRIK et al.: DIGITAL SINGLE LENS REFLEX CAMERA IDENTIFICATION FROM TRACES OF SENSOR DUST 549

Fig. 13. Canon EOS dust template created with three blank images with different f -numbers (F/13, F/22, F36).

DSLR camera (will be referred as Camera 3). Out of 110 im- To test source camera identification performance, 100 im-
ages, two different image sets were created. The first set con- ages were taken in different environments with different -num-
sisted of 15 images in which there was no apparent sky or ex- bers with each Canon and Nikon DSLR camera. To estimate
tensive flat region, and the second set consisted of 15 images the FP rate, 1000 images were taken with eight different digital
in which the sky was clearly visible. For each image set, three cameras (including Canon A80, Canon Rebel XT, Dimage Z3,
dust templates were generated from 5, 10, and 15 images, as de- Canon S2 IS, Cybershot, DSC-P72, DSC-S90, and EX-Z850).
scribed in Section III-B. The amount of dust in each dust tem- Then, the source camera identification procedure was performed
plate is given in Table V. Not surprisingly, the greatest number on these 100+1000 images for both Canon and Nikon dust tem-
of dust spots is achieved for 15 images with flat regions. To test plates. The identification confidence values for all 1100 images
the detection performances of these six templates, the proposed are given in Fig. 14 where the -axis represents image indices
camera identification scheme was tested using the rest of the 100 and the -axis represents the overall confidence values defined
images and 500 images taken with different DSLR and compact in (10). In the figures, the dot symbol corresponds to previously
cameras. The true-positive (TP) and false-positive (FP) rates for unseen images taken by the source DSLR camera. The dust
six different dust templates are given in Table VI. In the table, templates for Canon and Nikon DSLR cameras are comprised
the TP rate significantly increases as the amount of dust in the of 38 and 36 dust spots, respectively. The decision threshold
template increases with a small increase in FP (see Tables V and (threshold ) is set to fix FP probability at 0.002. The cor-
VI). responding TP rate and accuracy, where accuracy is defined as
It is seen from Table VI that high detection accuracy is pos- the ratio of all true detections to number of images, were com-
sible even with the five images used in template generation with puted as 0.610 and 0.963 for Nikon, and 0.920 and 0.991 for
smooth content. Nevertheless, to achieve such high accuracy Canon DSLR cameras. The TP rate for Nikon was significantly
with the images that do not contain any visible sky, the number smaller than the Canon image set due to the fact that the Nikon
of images used in template generation should be as high as set contained so many nonsmooth and complex images which
possible. made the decision more prone to error.

B. Case-I: Source Camera Available C. Case-II: Source Camera Unavailable

In this section, we assume that the DSLR source device is In this case, ten images taken with Nikon and Canon DSLR
available at hand. In experiments, we used Nikon D-50 and cameras were used to create the camera dust template and then
Canon EOS Digital Rebel cameras with 18–55 mm lenses. To the camera identification procedure was applied. In generating
introduce dust into the cameras, camera lenses were detached the camera dust template, images, which consist of mostly flat
while the cameras were powered up several times in an envi- regions, were used. Then, the camera identification scheme was
ronment where tiny particles, such as lint, hair, and dust, were applied to the same image sets which consist of im-
present. ages. The detection accuracy results obtained using dust tem-
Then, the camera dust templates of Nikon and Canon DSLR plates created from ten images with large smooth regions are
cameras were created by capturing three flat background pho- given in Fig. 15.
tographs at three different -numbers (i.e., F/13, F/22, and Due to the problem of creating an accurate dust template from
F/36). The template of the Canon DSLR is depicted in Fig. 13 a set of images, taken with uncontrolled conditions, with busy
along with one of the images used in template generation. In contents, the amount of dust in the dust templates decreased to
Fig. 13(b), the detected dust spots in the template are shown 4 from 36 for Nikon, and 10 from 38 for Canon, respectively.
as gray spots where the degree of darkness of the dust spots Since confidence metric increases with the number of dust spots
represents the number of hits in the dust template which (10), in Fig. 15, the range of confidence values decreases. The
was obtained as described in Section III-B. In Fig. 13(b), a small number of dust spots in the template makes it possible to
line-shaped lint particle is also detected as its size is very close achieve very low FP rates, with lower detection threshold. Thus,
to that of dust spots. the detection threshold was reset for the unavailable camera case
550 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO. 3, SEPTEMBER 2008

Fig. 15. Camera identification results (camera not available).

Fig. 14. Camera identification results (camera available).

1000 images from other digital cameras. The identification

(threshold ). For this new setting, the detection accu- performance for this image set is given in Fig. 17. It can be
racy for Nikon D50 decreased from 0.963 ( , seen from the figure that individual camera identification can
) to 0.951 ( , ). For Canon DSLR, be accomplished with 0.991 accuracy, 0.001 FP, and 0.910 TP
the detection accuracy decreased from 0.991 ( , rates for the given image sets for the same threshold (threshold
) to 0.985 ( , ) (see Figs. 14 ).
and 15). Although there is a small decrease in detection accu-
racy for the Nikon and Canon cameras, the FP rate is reduced to D. Robustness
zero despite a lower threshold value.
We evaluated the performance of the dust detection scheme
To make a more realistic experiment, the aforementioned
and source identification accuracy under two common types of
experiment was repeated using the image set (110 images of
processing.
camera 3) obtained from the web.8 Selecting the ten images
1) Downsizing: Since most dust spots appear with signif-
with the largest flat regions, the dust template of the DSLR
icant intensity degradations affecting a large group of pixels,
Sigma SD9 camera was obtained. The dust template and sample
they are not strongly affected from image resizing. To deter-
images used in template generation are depicted in Fig. 16.
mine the impact of downsizing on detection accuracy, 100 im-
In the template, there are 29 dust spots. The sky images in
ages of Canon EOS were downsized 50%. Then, the scheme was
the image set make it possible to reliably detect many dust
applied to original and downsized image sets. For the original
locations even though the actual camera is not available. The
Canon image set, 89 out of 100 images were detected correctly.
generated dust template is tested on the rest of 100 images and
This rate becomes 88 out of 100 for a 50% downsized image set.
8[Online]. Available: www.pbase.com/chucklantz/. It should be noted that in the camera identification scheme, all
DIRIK et al.: DIGITAL SINGLE LENS REFLEX CAMERA IDENTIFICATION FROM TRACES OF SENSOR DUST 551

TABLE VII
ROBUSTNESS TO JPEG COMPRESSION

Fig. 18. Effect of JPEG compression on dust-spot detection. The images are
Fig. 16. Dust template of camera 3 and the images (downloaded from the In- the outputs of NNC.
ternet) used in template generation.

The red points in the right figure show the falsely detected dust
spots as a result of JPEG compression.
The proposed identification scheme can be improved by rep-
resenting dust positions as nodes in a specific graph. This ex-
tension could make the proposed scheme more robust to geo-
metric/desynchronization attacks. However, for now, we leave
this extension as a future work.

V. CONCLUSION
In this paper, we have introduced a new source DSLR camera
identification scheme based on sensor dust traces. The location
and shape of dust specks in front of the imaging sensor and their
persistence make dust spots a useful fingerprint for DSLR cam-
eras. Although many DSLR cameras come with built-in dust
removal mechanisms, these hardware-based removal solutions
are not as effective as they claim to be. Besides, since most dust
spots are not visible or visibly irritating, most DSLR users ig-
nore them completely. To the our best knowledge, this is the first
Fig. 17. Identification results for 100 images downloaded from the Internet work in the literature which uses sensor dust spots for individual
(camera 3). camera identification. The efficacy of the proposed camera iden-
tification scheme is tested on higher than 1000 images from
different cameras. Experimental results show that the proposed
input images were resized to 800 533 resolution regardless of scheme provides high detection accuracy with very low false
input resolution. alarm rates. Our experimental tests also show that the proposed
2) JPEG Compression: To analyze the impact of compres- scheme is quite robust to JPEG compression and downsizing.
sion on the performance of source identification accuracy, 100 The biggest challenge in this research direction is the detection
images both from Nikon and Canon image sets and 500 images of dust spots in very complex regions and low -numbers.
from other digital cameras were compressed at JPEG quality 50.
The identification results are given in Table VII from which it ACKNOWLEDGMENT
can be seen that the proposed scheme is viable even under strong
JPEG compression. In the table, solely one Nikon image is iden- The authors would like to thank M. Pollitt at the University
tified better under JPEG compression. The NCC output for orig- of Central Florida for suggesting this line of research.
inal and compressed versions of that image is given in Fig. 18.
In the figure, it is seen that the JPEG compression increases the REFERENCES
number of local maxima exceeding the detection threshold in [1] D. L. M. Sacchi, F. Agnoli, and E. F. Loftus, “Changing history: Doc-
the NCC output. As a result, a dust spot which is not visible tored photographs affect memory for past public events,” Appl. Cognit.
in NCC output, corresponding to the original image, becomes Psychol., vol. 21, no. 8, pp. 1005–1022, Nov. 2007.
[2] H. Farid, Deception: Methods, Motives, Contexts and Consequences.
detectable after JPEG compression. However, at the same time, Stanford, CA: Stanford Univ. Press, 2007, ch. Digital Doctoring: Can
the number of false detections has also increased significantly. we trust photographs?.
552 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO. 3, SEPTEMBER 2008

[3] H. T. Sencar and N. Memon, “Overview of state-of-the-art in dig- [21] A. Krainiouk and R. T. Minner, “Method and system for detecting and
ital image forensics,” in Indian Statistical Institute Platinum Jubilee tagging dust and scratches in a digital image,” U.S. Patent 6 233 364
Monograph series titled Statistical Science and Interdisciplinary Re- B1, May 2001.
search. Singapore: World Scientific, 2008. [22] E. Steinberg , Y. Prilutsky, and P. Corcoran, “Method of detecting and
[4] T. V. Lanh, K.-S. Chong, S. Emmanuel, and M. S. Kankanhalli, “A correcting dust in digital images based on aura and shadow region anal-
survey on digital camera image forensic methods,” in Proc. IEEE Int. ysis,” pub. A1, Mar. 2005.
Conf. Multimedia Expo., 2007, pp. 16–19. [23] A. Zamfir, A. Drimbarean, M. Zamfir, V. Buzuloiu, E. Steinberg, and
[5] T.-T. Ng, S.-F. Chang, C.-Y. Lin, and Q. Sun, “Passive-blind image D. Ursu, “An optical model of the appearance of blemishes in dig-
forensics,” in Multimedia Security Technologies for Digital Rights, W. ital photographs,” Proc. SPIE, Digital Photography III, vol. 6502, pp.
Zeng, H. Yu, and A. C. Lin, Eds. New York: Elsevier, 2006. 0I1–0I12, Feb. 2007.
[6] G. Friedman, “The trustworthy digital camera: Restoring credibility to [24] E. Steinberg, P. Bigioi, and A. Zamfir, “Detection and removal of blem-
the photographic image,” IEEE Trans. Consum. Electron., vol. 39, no. ishes in digital images utilizing original images of defocused scenes,”
4, pp. 905–910, Nov. 1993. pub. A1, May 2007.
[7] P. Blythe and J. Fridrich, “Secure digital camera,” in Proc. Digital [25] J. Lewis, “Fast normalized cross-correlation,” Proc. Vision Interface,
Forensic Research Workshop, Aug. 2004, pp. 11–13. pp. 120–123, 1995.
[8] M. Kharrazi, H. T. Sencar, and N. Memon, “Blind source camera iden-
tification,” in Proc. IEEE Int. Conf. Image Processing, Oct. 2004, vol.
1, pp. 709–712.
[9] A. Swaminathan, M. Wu, and K. J. R. Liu, “Non intrusive forensic anal-
ysis of visual sensors using output images,” IEEE Trans. Inf. Forensics
Security, vol. 2, no. 1, pp. 91–106, Mar. 2007.
[10] Y. Long and Y. Huang, “Image based source camera identification Ahmet Emir Dirik received the B.S. and M.S.
using demosaicking,” in Proc. IEEE 8th Workshop Multimedia Signal degrees in electrical engineering from Uludag Uni-
Processing, Victoria, BC, Canada, Oct. 2006, pp. 4190–424. versity, Bursa, Turkey, and is currently pursuing the
[11] K. S. Choi, E. Y. Lam, and K. K. Y. Wong, “Source camera identifi- Ph.D. degree in signal processing at the Department
cation using footprints from lens aberration,” Proc. SPIE Digital Pho- of Electrical and Computer Engineering at the
tography II, vol. 6069, pp. 172–179, Feb. 2006. Polytechnic University, Brooklyn, NY.
[12] H. Gou, A. Swaminathan, and M. Wu, “Robust scanner identification His research interests include multimedia foren-
based on noise features,” Proc. SPIE Security, Steganography, Water- sics, information security, and data hiding.
marking of Multimedia Contents IX, vol. 6505, p. 65050, Feb. 2007.
[13] Z. J. Geradts, J. Bijhold, M. Kieft, K. Kurosawa, K. Kuroki, and N.
Saitoh, “Methods for identification of images acquired with digital
cameras,” Proc. SPIE Enabling Technologies for Law Enforcement
and Security, vol. 4232, pp. 505–512, Feb. 2001.
[14] K. Kurosawa, K. Kuroki, and N. Saitoh, “Ccd F ingerprint
method—Identification of a video camera from videotaped im- Husrev Taha Sencar received the Ph.D. degree in
ages,” in Proc. ICIP, Kobe, Japan, 1999, pp. 537–540. electrical engineering from the New Jersey Institute
[15] J. Lukáš, J. Fridrich, and M. Goljan, “Digital camera identification from of Technology, Newark, in 2004.
sensor noise,” IEEE Trans. Inf. Forensics Security, vol. 1, no. 2, pp. Currently, he is a Postdoctoral Researcher with
205–214, Jun. 2006. the Information Systems and Internet Security Labo-
[16] M. Chen, J. Fridrich, and M. Goljan, “Digital imaging sensor identi- ratory of the Polytechnic University, Brooklyn, NY.
fication (further study),” Proc. SPIE Security, Steganography, Water- His research interests are the security of multimedia
marking of Multimedia Contents IX, vol. 6505, p. 65050, Feb. 2007. and communications.
[17] M. Chen, J. Fridrich, M. Goljan, and J. Lukáš, “Source digital cam-
corder identification using sensor photo response non-uniformity,”
Proc. SPIE Security, Steganography, Watermarking of Multimedia
Contents IX , vol. 6505.G, p. 65051, 2007.
[18] N. Khanna, A. K. Mikkilineni, G. T. C. Chiu, J. P. Allebach, and E. J.
Delp, “Scanner identification using sensor pattern noise,” Proc. SPIE
Security, Steganography, Watermarking of Multimedia Contents IX, Nasir Memon is a Professor in the Computer Science
vol. 6505, p. 65051, Feb. 2007. Department at the Polytechnic University, Brooklyn,
[19] A. E. Dirik, T. H. Sencar, and M. Nasir, “Source camera identification NY.
based on sensor dust characteristics,” in Proc. IEEE Workshop Signal He is the Director of the Information Systems and
Processing Applications for Public Security and Forensics, Apr. 2007, Internet Security (ISIS) Lab at Polytechnic Univer-
pp. 1–6. sity. His research interests include data compression,
[20] T. Gloe, E. Franz, and A. Winkler, E. J. Delp III and P. W. Wah, Eds., computer and network security, digital forensics, and
“Forensics for flatbed scanners,” in Proc. Security, Steganography, and multimedia data security.
Watermarking of Multimedia Contents IX., Feb. 2007, vol. 6505, p.
65051.