NOVATEUR PUBLICATIONS

International Journal of Research Publications in Engineering and Technology [IJRPET]

ISSN: 2454-7875
VOLUME 3, ISSUE 3, March-2017
ADVANCED FILTERING TECHNIQUES IN IMAGE PROCESSING : A SURVEY
MISS. DESHPANDE MAYURI
Department of Electronics Engineering, Bharat Ratna Indira Gandhi College Of Engienering, Kegaon, Solapur, India

MR. KOLKURE V.S.

Department of Electronics Engineering, Bharat Ratna Indira Gandhi College Of Engineering, Kegaon, Solapur, India

ABSTRACT: separated by sharp edges, this is called edge preserving

Nowadays, in the field of image processing so many regularization[1].
new technologies are found out and for various
purposes such as filtering, edge smoothing, edge A. REGULARIZED IMAGE RECONSTRUCTION:
sharpening, zooming, enhancing, fusing, noise and Char bonnier et al proposed method of
blur removing . But these primitive and conventional regularization of images[1]. The image and the auxiliary
methods bring some noises and halo artifacts with variables are either assumed as N×N two dimensional
them as well as they need more time for processing. fields indexed by i(row number) and j(column number)
The algorithms used in these methods are very large or denoted as N2 × 1 exicographically ordered vectors
and complicated. This method mainly focuses on indexed by k. The index transformation between the two
filtering of images along with image enhancing, representations is,
image fusion, haze removal and detail enhancement. k = i × N + j [1].
Weighted Guided Image Filter uses two types of In the field of computed imaging, the observed
filtering algorithm: First is Global Filtering and data is related to the original image by the following
second is Local Filtering. Therefore, it makes relation :
filtering process easy, less complicated and less time p=Rf+η (1)
consuming. where,
KEYWORDS: Haze removal, exposure fusion, p is observed data, f is original image, R is block Toeplitz
weighted guided image filtering, detail and represents the point spread function of the imaging
enhancement, edge aware weighting, and halo system[2], η is white Gaussian noise[14]. Before the
artifacts. evolution of edge-aware techniques of image filtering,
images are restored by using equation(1). This equation
I. INTRODUCTION: preserves edges therefore it is called as deterministic
Digital image processing is becoming very edge preserving regularization. But this reconstructed
crucial tool in every field of science. The field may be image contains noise and halo artifacts. Sharp edges
medical imaging, film making or study of earth, study of involved in the image create obstacle in the smooth
cosmic atmosphere. This requires better image clarity, vision. Therefore, smoothing and filtering is very
smooth edges, enhanced vision [14]. This Weighted essential for clear and enhanced vision.
Guided Image Filter deals all of these and basically edge
preserving filtering. B. FILTERING TECHNIQUES:
The method of noise removal is classified into As mentioned before, performing filtering and
two types : Global filters and second is Local filtering. In smoothening is very important for clear vision. There are
global filters, the performance criteria consist of a data some classifications of filtering techniques such as
term and a regulation term. Local filters include bilateral bilateral filtering, trilateral filtering, gradient
filters, trilateral filters and gradient filters [4],[5]. minimization for smoothing. These are given in brief as
Here, edge aware weighting is introduced. In below:
edge aware weighting, edges play very important role in
vision of objects [6]. Heavy weights are assigned to 1. IMAGE SMOOTHING VIA L0 GRADIENT
pixels at edges than those pixels which are at flat areas. MINIMIZATION:
Due to this proposed weighting technique, sharp edges This method is related to edge-preserving
are saved and smoothened for better resolution and smoothing which aims to maintain and improve
vision[8]. prominent set of edges. In this method, a sparse gradient
counting scheme in optimization framework is
II. RELATED WORKS ON EDGE-AWARE WEIGHTING proposed[3]. The different techniques such as 1D and 2D
TECHNIQUES: smoothing. These both techniques are explained in brief
In this section, various edge preserving as below :
techniques are discussed. This section establishes base
for the proposed work of edge-aware weighting A. 1D smoothing
technique. While dealing with the reconstruction of an In 1D smoothing, highest contrast edges are
image, it is necessary to regularize the solution. A simple enhanced and smoothing is done in global manner. L. Xu
regularization process assumes that pixels in the image et al had given the method of 1D smoothing where input
are equally distributed. A practical model of an image discrete signal is denoted by g and its smoothed signal is
assumes that images are made up of smooth regions
148 | P a g e
 NOVATEUR PUBLICATIONS
International Journal of Research Publications in Engineering and Technology [IJRPET]
ISSN: 2454-7875
VOLUME 3, ISSUE 3, March-2017
denoted by f. This method counts amplitude changes trilateral filter given by P.Choudhary and Tumbline
and is given as, consists of two modified bilateral filters. Its novel
c(f) = #{p | | fp − fp+1| = 0} , contributions given by them are :
where p and p + 1 index neighboring samples. | fp − (a) Tilting : Filter window is skewed by bilaterally
fp+1| is a gradient w.r.t. p in the form of forward smoothed image gradient vector.
difference. #{} is the counting operator, outputting the (b) Adaptive region growing : The domain automatically
number of p. adapts to local image features.
B. 2D formulation (c) One parameter : Trilateral filter uses only one
In 2D formulation, the color difference between parameter for filtering.
two pixels is calculated along x and y directions. The These three novel contributions come together
gradient magnitude for color images is given as the sum to form the trilateral filter with adjustable parameters.
of gradient magnitude in rgb. Farbman et al has given the Trilateral filters give high noise reduction, better vision,
solver for 2D formulation [2008]. mesh smoothing, HDR tone mapping[11],[13]. Trilateral
filtering has the ability to separate details from noisy
2. BILATERAL FILTERING: input image. These bilateral and trilateral filters are used
Bilateral filtering preserves edges and smoothes in weighted image guided filter algorithm.
images in the same way as that of human perception[5].
Bilateral filtering is a non iterative and simple method of 4. REAL TIME EDGE PRESERVING:
filtering. This scheme of bilateral filtering can be carried Edge preserving image processing is a new
out by simple neuron like devices which perform their technique of filtering the image[6]. In this technique,
operation once per image. In bilateral filtering, three edges of the images are preserved and smoothed with
bands are filtered separately because different bands the help of bilateral grid. J.Chen and team has given the
have different levels of contrast[10]. technique for fast filtering and edge preserving.
Therefore, bilateral filters can operate three The bilateral grid used in this technique is
bands at once. Bilateral filters do the same work in the defined as, a 3D array that combines two-dimensional
image range which traditional filters do their work in its spatial domain with one-dimensional range domain. The
domain. The combination of range filtering and domain bilateral grid is first appeared in Paris and Durand's fast
filtering gives excellent filtration and hence called as bilateral filter [2006].
bilateral filtering. In bilateral filtering, a low pass filter is Bilateral grid is sampled continuously in each
applied and then the value of a pixel is replaced by an domain, let us consider, the sampling rate of the spatial
average and similar nearby pixel values. axes as Ss and the sampling rate of range axis as Sr. Then
Suppose, a low pass filter is applied to image f(x) bilateral grid given by J. Chen and team is as follows,
which produces output image, (a) Initialization :
∞ ∞
h(x) = kd (x) ∫−∞ ∫−∞ f(ξ) c(ξ, x)dx
-1 (2) For all grid nodes (i, j, k), Г(i, j, k) = (0,0)
where, c(ξ,x) denotes the geometric closeness between where, I is the input image and Г is the constructed
nearby centre x and neighborhood point ξ. The range bilateral grid.
filter is defined as, (b) Filling :
∞ ∞
h(x) = kr-1(x) ∫−∞ ∫−∞ f(ξ)s[f(ξ), f(x)]dξ (3) For each pixel at point (x,y),
Г ( [x/Ss] , [y/Sr] , [I(x,y)/Sr] ) = ( I(x,y) , 1),
where, function 's' operates in the range of the image where [.] is the closest integer operator.
function 'f' . Then, combined filtering is defined as, (c) Processing :
∞ ∞
h(x) = ∫−∞ ∫−∞ f(ξ) c(ξ, x) s[f(ξ), f(x)] dx (4) Any function f which is able to take 3D function
with the normalization, as input can be applied to this grid. The grid processing
∞ ∞
k(x) = ∫−∞ ∫−∞ c(ξ, x) s[f(ξ), f(x)]dξ (5) stage of bilateral filter is convoluted by a 3D Gaussian
This combined filter in equation (5) is called as bilateral kernel. After processing, the next stage is slicing. The
filter. This bilateral filter is used for noise reduction, grid is sliced using the input image I to obtain final
removal of halo artifacts and blur. A grid can be added in output which in 2D form.
the bilateral filter for better enhancement and clear This bilateral grid is used to enhance the
vision. These both techniques are involved in the contrast of the images which is an essential tool in High
invention of trilateral filters. Dynamic Range images.
The contrast of the image can be enhanced by
3. TRILATERAL FILTERING: adjusting the histogram of that image. This real time
Trilateral filtering is advanced type of filtering edge preserving technique can be applied for the
as compared with bilateral filtering. It is a single-pass purpose of progressive abstraction which is very
non linear filter used in edge preserving algorithms for important in the medical field nowadays.
better vision and gives stronger noise reduction. This
trilateral filter is built from bilateral filter of C. Tomasi 5. GUIDED IMAGE FILTER:
and R. Manduchi [5]. Conventional methods of filtering include Linear
Trilateral filter requires only one user set Translation Invariant such as Gaussian, Laplacian and
parameters and filters the image in only one pass[4]. It Sobel filters[9],[14]. But there are some drawbacks of
does not require an iterative solver for filtering. The these filters. A new technique is discovered to overcome

149 | P a g e
 NOVATEUR PUBLICATIONS
International Journal of Research Publications in Engineering and Technology [IJRPET]
ISSN: 2454-7875
VOLUME 3, ISSUE 3, March-2017
these drawbacks and it is Guided Image Filter[7]. Guided 6) C. Tomasi and R. Manduchi, “Bilateral filtering for
image filter has good edge preserving properties and gray and color images,” in Proc. IEEE Int. Conf.
does not involve reverse gradient artifacts. Various edge Comput. Vis., Jan. 1998, pp. 836–846.
preserving techniques are classified into explicit 7) J. Chen, S. Paris, and F. Durand, “Real-time edge-
weighted average filters and implicit weighted average aware image processing with the bilateral grid,” ACM
filters. Guided image filter have combined properties of Trans. Graph., vol. 26, no. 3,
these two. 8) pp. 103–111, Aug. 2007.
A general linear translation filtering process is 9) K. He, J. Sun, and X. Tang, “Guided image filtering,”
known as guided image filter and involves guidance IEEE Trans. Pattern Anal. Mach. Intell., vol. 35, no. 6,
image, filtering input image and output image. And hence pp. 1397–1409, Jun. 2013.
it is widely used in advanced filtering techniques. 10) Z. Farbman, R. Fattal, D. Lischinski, and R. Szeliski,
“Edge-preserving decompositions for multi-scale tone
III. CONCLUSION: and detail manipulation,” ACM Trans. Graph., vol. 27,
In this paper, various techniques of filtering, no. 3, pp. 249–256, Aug. 2008.
edge preserving, smoothing, noise removing, enhancing 11) P. Pérez, M. Gangnet, and A. Blake, “Poisson image
and compression are discussed in brief[8]. Weighted editing,” ACM Trans. Graph., vol. 22, no. 3, pp. 313–
guided image filtering is the most advanced technique of 318, Aug. 2003.
noise removal, image fusion, contrast adjustment[15]. 12) Z. Li, J. Zheng, Z. Zhu, S. Wu, and S. Rahardja, “A
Weighted guided image filter uses primitive techniques bilateral filter in gradient domain,” in Proc. Int. Conf.
of image filtering and combines them for better results. Acoust., Speech Signal Process.,
This survey shows how weighted guided image 13) Mar. 2012, pp. 1113–1116.
filter is the better option for image processing[7]. 14) F. Durand and J. Dorsey, “Fast bilateral filtering for
Conventional techniques of image processing require the display of high dynamic- range images,” ACM
more time for processing and large calculations, Trans. Graph., vol. 21, no. 3, pp. 257–266, Aug. 2002.
equations. But, weighted guided image filter has less 15) B. Y. Zhang and J. P. Allebach, “Adaptive bilateral
running time and less complicated algorithm as filter for sharpness enhancement and noise removal,”
compared with primitive techniques. Therefore, it is IEEE Trans. Image Process., vol. 17, no. 5, pp. 664–
becoming a widely used tool in the field of image 678, May 2008.
processing. 16) C. C. Pham, S. V. U. Ha, and J. W. Jeon, “Adaptive
guided image filtering for sharpness enhancement
IV. ACKNOWLEDGMENT: and noise reduction,” in Advances in Image and
First and foremost, I would to thank my Video Tec-
academic advisor Mr. V . S. Kolkure for guiding and 17) hnology. Berlin, Germany: Springer-Verlag, 2012.
helping me with his great subject acknowledgment. 18) R. C. Gonzalez and R. E. Woods, Digital Image
During my tenure, he contributed to a rewarding Processing. Upper Saddle River, NJ, USA: Prentice-
experience by giving me intellectual freedom in my Hall, 2002.
work, engaging me in new ideas and demanding high 19) Z. Li, J. Zheng, Z. Zhu, and S. Wu, “Selectively detail-
quality work in all my endeavors. enhanced fusion of differently exposed images with
A number of people have generally given moving objects,” IEEE Trans. Image Process., vol. 23,
helpful comments, during the preparation of the no. 10, pp. 4372–4382, Oct. 2014.
manuscript and cooperation. I also thank our Principal
for showing positive response to encourage me. I also
thank to all direct and indirect help provided by the staff
and the entire batch mates for their enthusiasm and
great ideas.

REFERENCES:
1) P. Charbonnier, L. Blanc-Feraud, G. Aubert, and
M.Barlaud, “Deterministic edge-preserving
regularization in computed imaging,” IEEE Trans.
2) Image Process., vol. 6, no. 2, pp. 298–311, Feb. 1997.
3) A.K. Jain. The fundamentals of digital image
processing, Engle wood cliffs, NJ : Prentic hall 1989.
4) L. Xu, C. W. Lu, Y. Xu, and J. Jia, “Image smoothing via
L0 gradient minimization,” ACM Trans. Graph., vol.
30, no. 6, Dec. 2011, Art. ID 174.
5) P. Choudhury and J. Tumblin, “The trilateral filter for
high contrast images and meshes,” in Proc.
Eurograph. Symp. Rendering, pp. 186–196, 2003.

150 | P a g e