SOM A Computer Vision Technique For Face Sketch Featured Database

International Journal on Recent and Innovation Trends in Computing and Communication
ISSN 2321 8169

355 358
Volume: 1 Issue: 4
___________________________________________________________________________
SOM: A Computer Vision Technique for Face Sketch Featured Database

Vineet Srivastava
Dehradun Institute of Technology
Dehradun, Uttarakhand
[email protected]
Abstract Self-Organizing Maps (SOMs) found to be an improved data management computer vision technique used for the
closed matching of face vs sketch identification system based on neural network of untrained input images with trained database of
images. Parameters for the SOM neural network are selected to be a minimum and maximum point for each row on a vector of
training database. In this paper 64 minimum and 64 maximum pixel intensity values selected altogether using 8x8 image masking
technique. Further for the design of SOM a set of 25 uniform image data used to create 5 different classes of a face image like left
eye, right eye, nose, frontal face and lips for the training database. All the preprocessing for the image enhancement is done in the
MATLAB software.
Keywords-SOM; Masking; 2D-DCT; Computer Vision; Neural Network
_____________________________________________________*****______________________________________________________
I. INTRODUCTION
Human communication has two main aspects auditory
(verbal) and visual (non-verbal). Non-verbal communication
like facial expression, body movements, and physiological
reactions provides significant information regards the state of
the person. Computer vision aims to duplicate the human
vision by electronically perceiving and understanding an
image. Computer vision techniques use the results of
artificial intelligence, pattern recognition, mathematics,
computer science, psycho-physiology and other scientific
areas.
II. KOHONEN SELF-ORGANIZING MAPS
Self-Organizing Maps (SOM) was introduced by a
Finnish Professor, Teuvo Kohonen in 1982, thus SOMs are
also sometimes known as Kohonen Maps. SOM are subtype
of artificial neural network. They are trained using unsupervised learning to produce low dimensional
representation of the training samples while preserving the
topological properties of the input space. Thus SOM are
reasonable for visualizing low-dimensional views of highdimensional data.
Self-Organizing Maps are a single layer feed forward
network where the output syntaxes are arranged in low
dimensional grid i.e. 2D or 3D. Each input is connected to all
output neurons attached to every neuron there is a weight
vector with the same dimensionality as the input vectors. The
number of input dimensions is usually a lot higher than the
output grid dimension. A Self-organizing map is shown in
below Figure 1.
Figure 1: SOM graphical view

A. Network Architecture
Self-organizing maps are single layer feed-forward
networks where the output syntaxes are arranged in low
dimensional (usually 2D or 3D) grid. Each input is
connected to all output neurons. Attached to every neuron
there is a weight vector with the same dimensionality as the
input vectors. The number of input dimensions is usually a
lot higher than the output grid dimension. SOMs are mainly
used for dimensionality reduction rather than expansion.
The architecture for a simple self organizing map is shown
in Figure 2.
Figure 2: Network Architecture of SOM

The input vector p is the row of pixels of the image. The
||ndis|| box in the Figure 4.3.2.1.2 accepts the input vector p
and the input weight matrix IW1, 1 produces a vector having
S1 elements. The elements are the negative of the distances
between the input vector and vectors IW1, 1 formed from the
rows of the input weight matrix. The competitive transfer
function accepts a net input vector for a layer and returns
neuron outputs of 0 for all neurons except for the winner,
the neuron associated with the most positive element of net
355
IJRITCC | MAR 2013, Available @ http://www.ijritcc.org
___________________________________________________________________________
ISSN 2321 8169

355 358
Volume: 1 Issue: 4
___________________________________________________________________________
input n1. The winners output is 1. The neuron whose weight
vector is closest to the input vector has the least negative net
input and, therefore, wins the competition to output a 1.
Thus the competitive transfer function produces a 1 for
output element a1i corresponding to i* the winning neuron.
All other output elements in a1 are 0.
A self-organizing feature map network identifies a
winning neuron using the same procedure as employed by a
competitive layer. However, instead of updating only the
winning neuron, all neurons within a certain neighborhood
of the winning neuron are updated using the Kohonen rule.
III.
E. 2D-DCT
The two-dimensional discrete cosine transform (2DDCT) used for image processing. The 2D-DCT resembles
the 1D-DCT transform because it is a separable linear
transformation. For example, in an n x m matrix, S, the 2DDCT is computed by applying it to each row of Sand then to
each column of the result. Figure 3 shows the generic 2DDCT architecture of an N x M input image.
DISCRETE COSINE TRANSFORM (DCT)
A. Overview
The discovery of Discrete Cosine Transform (DCT) in
1974 was an important achievement for research community
working on image compression. It is a lossy data
compression technique virtually used image processing.
Compression standard like JPEG for compression of still
images employ the basics of the DCT.
B. Background
The DCT algorithm similar to Fast Fourier Transform
converts the data pixels into the sets of frequencies. The first
frequencies in the set are the most meaningful than the latter
i.e. least. To compress the data the least meaningful
frequencies are stripped away based on the allowable
resolution loss. DCT operates on a function at a finite
number of discrete data points.
C. Definition
The DCT is regarded as a discrete-time version of the
Fourier cosine series. Hence it is considered as a Fourierrelated transform similar to the Discrete Fourier Transform
(DFT) using only real numbers.
D. 1D-DCT
The Discrete Cosine Transform is a linear invertible
function or an N x N square matrix like:
F: RN -> RN
where, R denotes set of real numbers
Mathematically,1D-DCT
length N can be defined as:
Figure 3: 2D-DCT
IV. DESIGN AND IMPLEMENTATION
In Computer Vision, detecting a Face Sketch in a digital
image involves segmentation, extraction and verification of
faces and possibly facial features like Left-Eye, Right-Eye,
Nose, Lips and frontal face from an uncontrolled
background.
A. Face Sketch Recognition
Face Sketch recognition algorithm can be classified into
two main categories i.e. generative and discriminative
approaches.
1) Genrative Approach: To model a digital image in
terms of sketches and then match it with the query sketch or
vice-versa.
Wang and Tang proposed Eigen transformation based
approach to transform a digital photo into sketch before
matching. In another approach, they presented an algorithm
to separate shape and texture information and applied
Bayesian classier for recognition. Liu et al. proposed nonlinear discriminative classier based approach for
synthesizing sketches by preserving face geometry. Li et al.
matched sketches and photos using a method similar to the
Eigen-transform after converting sketches to photos.
Recently, Wang and Tang proposed Markov Random Fields
based algorithm to automatically synthesize sketches from
digital face images and vice-versa.
X[k] of a sequence x[n]of
Each element of the transformed list X[k] in above

equation is the inner dot product of the input list x[n] and a
basis vector. Constant factors are chosen so the basis vectors
are orthogonal and normalized. The DCT can be written as
the product of a vector (input list) and the N x N are
orthogonal matrix whose rows are the basis vectors.
2) Discriminative Approach: This performs feature

extraction and matching using the given digital image and
sketch pair and do not generate the corresponding digital
image from sketches or the sketch from digital images.
Uhl and Lobo proposed photometric standardization of
sketches to compare it with digital photos. The sketches and
photos were geometrically normalized and matched using
Eigen analysis. Yuen and Man used local and global feature
measurements to match sketches and mug-shot images.
Zhang et al. compared the performance of humans and
PCA-based algorithm for matching sketch-photo pairs with
356
___________________________________________________________________________

Volume: 1 Issue: 4
ISSN 2321 8169

355 358
___________________________________________________________________________
variations in gender, age, ethnicity and inter-artist
variations. They also discussed about the quality of sketches
in terms of artists skills, experience, exposure time, and
distinctiveness of features. Similarly, Nizami et al. analyzed
the effect of matching sketches drawn by different artists.
Klare and Jain proposed a Scale Invariant Feature
Transform (SIFT) based local feature approach where
sketches and digital face images were matched using the
gradient magnitude and orientation within the local region.
Bhatt et al. extended Uniform Local Binary Patterns to
incorporate exact difference of gray level intensities to
encode texture features in sketches and digital face images.
Klare et al. extended their approach using Local Feature
Discriminant Analysis (LFDA) to match forensic sketches.
In their recent approach, Klare and Jain proposed a
framework for heterogeneous face recognition where both
probe and gallery images are represented in terms of
nonlinear kernel similarities. Recently, Zhang et al.
proposed an information theoretic encoding band descriptor
to capture discriminative information and random forest
based matching to maximize the mutual information
between the sketch and the photo. [7]
B. Face Image Pre-processing
The programming language used to design and
implement the forensic face sketch identification system
code is MATLAB. The reason for using MATLAB in this
project is due to its Neural Network and Image Processing
toolbox that helped to obtain an efficient code.
Face image processing consists of the following steps:
Data Gathering
Import Faces Images to MATLAB
Image Resize in MATLAB
Featured Cropping
1.
Data Gathering
Face images of different were taken from CUHK

database which are stored under uniform light conditions
and frontal position with similar dimensions. Figure 4
shows some of the original face vs sketch pictures.
Face images were then preprocessed as:
Image conversion from RGB color to 8-bit
grayscale
Image resizing to 512 x 512 pixels
Figure 4: Face vs Face Sketches Images from the CUHK

Database
2. Import of Face Images to MATLAB
All face images were preprocessed in Adobe
Photoshop, then they were imported into MATLAB. The
MATLAB command imread was used to load pictures into
the workspace.
3. Image Re-size in MATLAB
After all face images were imported into MATLAB,
they were resized further from 512 x 512 pixels to 8 x 8
pixels. For this purpose the MATLAB command imresize
was used to resize the imported pictures.
4. Block Preparation
For the block preparation an image is divided into
individual blocks. A block consists of 8x8 pixels. Images
are divided into blocks because each block is treated
individually (compression steps are applied onto individual
blocks, not onto the image as a whole). Figure 5 illustrates
block preparation by dividing an image into a block of 8x8
pixels.
357
___________________________________________________________________________
ISSN 2321 8169
355 358
Volume: 1 Issue: 4
___________________________________________________________________________
1
1
0
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0 ];
Figure 7: DCT Masking Matrix
Figure 5: Image Block of 8x8 Pixels

5. Quantization
The block of 8x8 DCT coefficients are divided by an
8x8 quantization table. In quantization the low DCT
coefficients of the high frequencies are discarded. Thus,
quantization is applied to allow further compression of
entropy encoding by neglecting insignificant low
coefficients.
6. Featured Cropping
After all face images were resized, they were cropped as
Left-Eye, Right Eye, Nose, Lips , frontal face with hair
removes and saved under different filenames, using the
MATLAB imwrite command.
Left Eye Right Eye Nose Frontal Face

Figure 6: Featured Class Databases
V.
Lips
FACE IMAGE CLASSES DCT
After all featured cropped face images were resized to 8 x

8 pixels and saved, the next step was to compress them by
applying the 2D blocked DCT. When the 2D DCT is
applied with a mask, high-coefficients are in the image
discarded. Then the 2D IDCT is applied to regenerate the
compressed image, which is blurred due to loss of quality
and also smaller in size. To find a technique to apply the 2D
DCT to a face image, the MATLAB help was searched in
the Image Processing Toolbox. A program for 2D DCT
image compression in MATLAB help was found with its
source code.
The source code found from the MATLAB image
processing toolbox help was used to DCT all face images,
after few modifications. Before DCT compression, the
image data of the resized images needed to be converted
into the double format. This was achieved by using the
MATLAB double command. The mask used for the 2DDCT was of 8 coefficients out of 64. Figure 7 shows the
DCT masking matrix.
mask = [1 1 1 1 0 0 0 0
Tthe masking host was set to 8 coefficients and all

featured face/face sketch images were compressed using the
DCT. The newly compressed face images were saved under
a different filename.
FUTURE SCOPE AND CONCLUSION
In this paper I demonstrated the preprocessing techniques
for face vs face sketch recognition based system using SOM
neural network under a approach of computer vision.
Generally, image based data have large numbers of pixel
values therefore to perform any computation on those pixel
values a mind frame is needed for a particular region of
interest because images intensity and color information are
mostly based on very less in deviation with the nearby pixel
values. In the SOM algorithm, we calculates the deviation
against the higher side and lower side of images data and
meanwhile the closed pixels values discarded to save our self
from duplicate values.
SOM is a fast technique based on masking of 8x8 matrix
data to prepare a vector of 64 values which will be further
used to match against lower and higher side values of
claimed face sketch image. In future scope of this paper one
can apply a pattern recognition technique based on all
masking pixel values to make a set of different featured base
database that could be used to faster matching of the claimed
image.
REFERENCES
[1] http://www.willam
ette.edu/~gorr/classes/cs449/Unsupervised/SOM.html
Kohonens Self-Organizing Map.
[2]
http://www.ai-junkie.com/ann/som/som1.html,
Kohonens Self
Organizing Feature Maps, Tutorial.
[3] Anil K. Jain, Brendan Klare, and Unsang Park, Face
Matching and Retrieval in Forensics Applications,
IEEE Multimedia Forensic, Security and Intelligence,
January-March, 2012.
[4] Himanshu S. Bhatt, Samarth Bharadwaj, Richa,
Mayank Vatsa, Memetic Approach for Matching
Sketches with Digital Face Images, Submitted to IEEE
Transactions on PAMI.
[5] http://dmr.ath.cx/gfx/dct/, The 2D Discrete Cosine
Transform and ImageCompression.
[6]
http://mmlab.ie.cuhk.edu.hk/facesketch.html
358
___________________________________________________________________________

Volume: 1 Issue: 4
ISSN 2321 8169

355 358
___________________________________________________________________________
359
___________________________________________________________________________

SOM A Computer Vision Technique For Face Sketch Featured Database

Uploaded by

Copyright:

Available Formats

SOM A Computer Vision Technique For Face Sketch Featured Database

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SOM A Computer Vision Technique For Face Sketch Featured Database

Uploaded by

Copyright:

Available Formats

International Journal on Recent and Innovation Trends in Computing and Communication

ISSN 2321 8169