A Comparative Analysis of Digital Filters for

Image Processing Applications

Varsha.K.1,Dr. Vijaya Lakshmi D.2(Assistant Professor)
Department of VLSI and Embedded Systems, Bangalore Institute Of Technology, Bangalore, India

Abstract- Processing of a digital image by the use of computer description out)

algorithm is usually referred as digital image processing. An image may be considered to contain sub-images
Image filtering is one of the very useful techniques in image sometimes referred to as regions-of-interest, ROIs, or simply
processing and computer vision. It is used to remove some regions. This concept reflects the fact that images frequently
unwanted signals capture the image and termed as noise . contain collections of objects each of which can be the basis
Digital filters are an integral part of signal processing which for a region. In a sophisticated image processing system it
help us to remove the frequencies which are of less should be possible to apply specific image processing
importance or considered unwanted, also termed as noise operations to selected regions. Thus one part of an image
sometimes. Thus the digital image filtering technique is (region) might be processed to suppress motion blur while
frequently used to eliminate noise. Generally image filters another part might be processed to improve color rendition.
follows software approach in systems. But hardware Sequence of image processing: Most usually, image
implementation prefered in comparison with software processing systems require that the images be available in
implementation for better processing speed. With a boost in digitized form, that is, arrays of finite length binary words.
the VLSI technology hardware implementation has given a For digitization, the given Image is sampled on a discrete grid
better choice. This project presents the design of a new and each sample or pixel is quantized using a finite number of
hardware accelerator, filtering the input data using Gabor bits. The digitized image is processed by a computer. To
and Box functions and dedicated to image processing. The display a digital image, it is first converted into analog signal,
proposed design obtains a great reduction in terms of which is scanned on to a display.
resources compared to other state-of-the-art
implementations. In this work, a comparative study on FPGA In fact, each captured image contains noise. Due to various
implementation of both the Gabor filter and Box filter is interferences, noise, the image definition gets bad influence.
presented. Simulation and Synthesis is done using Xilinx At the same time, noise making the image blurred. The bad
ISE design suite. Verilog code will be used as a description condition was submerged fully. It gives analysis big difficulty.
language for mapping algorithm in VLSI and hardware Therefore, people need to suppress unwanted noise to improve
implementation on SPARTAN-3 FPGA. image quality. For digital image noise reduction, the basic
filtering algorithms are used. It includes mean filtering,
Keywords-Image Processing,Digital Filters,Gabor Filter,Box median filtering, Gaussian filtering, bilateral filtering.
Filter.
The main Problem encountered in Image processing is:
I. INTRODUCTION
 It suffers from halo-artifacts found along edges where
Impression is more by an image rather than thousands of words the intensities in the image vary drastically.
as stipulated by the statement “A picture is worth thousands”.  It results in problems of performance reduction with the
Although ‘pattern recognition’ and ‘image processing’ have a image quality or optimal threshold selection.
lot in common, yet they were developed as separate
disciplines. Two broad classes of techniques, viz., processing Motivated by the requirement of image filtering in a wide
and analysis have evolved in the field of digital image range of applications, the better filtering quality in edge
processing and analysis. Processing of an image includes preservation by the gradient domain guided image filter
improvement in its appearance and efficient representation. So algorithm and the lack of hardware implementation of
the field consists of not only features extractions, analysis and gradient domain guided image filter, we propose a VLSI
recognition of images, but also coding, filtering, enhancement architecture for the gabor filter and box filter with several
and restoration. novel techniques to address the challenges. The proposed
filter designs are presented in detail in the following sections.
In imaging science, image processing is any form of
signal processing for which the input is an image, such as This paper is organized as follows. Section II presents the
a photograph or video frame; the output of image literature survey based on the previous work carried out
processing may be either an image or a set of related to various applications on gabor filter, box filters,
characteristics or parameters related to the image. Most space filters, Bilateral Filter, Linear spatial Filter, guided
image-processing techniques involve treating the image as image filter and so on. Section III presents the details of the
a two-dimensional signal and applying standard signal-
proposed system implementation of Gabor and Box filters.
processing techniques to it. Image processing usually
refers to digital image processing, but optical and analog Section IV presents the simulation and hardware
image processing also are possible. An image defined in the implementation results carried out using Xilinx and FPGA
“real world” is considered to be a function of two real Platform. Finally, conclusion and the future scope of the
variables, for example, a(x,y) with a as the amplitude (e.g. project is presented.
brightness) of the image at the real coordinate position
(x,y).The goal of this manipulation can be divided into
three categories: II. LITERATURE SURVEY
 Image Processing (image in to image out)
 Image Analysis (image in to measurements out) Convolutional Neural Networks (CNN) are being increasingly
 Image Understanding (image in to high-level used in computer vision for a wide range of classification and
recognition problems. However, training these large networks noiseless and fairly good results in every case, which assesses
demands high computational time and energy requirements; the high robustness of this technique. The performance of the
hence, their energy-efficient implementation is of great proposed method[3] is compared with some classical edge
interest. Syed ShakibSarwar et al.[1] reduced the training detection techniques (e.g. Sobel, Prewitt, Roberts, Laplacian
complexity of CNNs by replacing certain weight kernels of a of Gaussian, Zero-Cross, Canny etc.) and methods reported in
CNN with Gabor filters. The convolutional layers use the some papers. Even though this method produces better results,
Gabor filters as fixed weight kernels, which extracts intrinsic it fails to image-shadow elimination, so the next venture will
features, with regular trainable weight kernels. This be definitely overcome this limitation and use genetic
combination creates a balanced system that gives better algorithm to make the system more efficient.
training performance in terms of energy and time, compared to
the standalone CNN (without any Gabor kernels), in exchange Box filter is a fundamental filter in image processing.
for tolerable accuracy degradation. They show that the However, it cannot preserve edges or corners. Yuanhao Gong
accuracy degradation can be mitigated by partially training the et al.[4] presents a simple but novel method that can make box
Gabor kernels, for a small fraction of the total training cycles. filter both edge and corner preserving. More specifically, they
They evaluated the proposed approach on 4 benchmark combine the box filter with sub-window regression to achieve
applications. Simple tasks like face detection and character this task. This filter inherits some properties from box filter,
recognition (MNIST and TiCH), were implemented using such as O(1) running time with respect to the window radius.
LeNet architecture. While a more complex task of objet After analyzing its parameters, they show its corner and edge
recognition (CIFAR10) was implemented on a state‐of-the‐art preserving property on real images and compare it with
deep CNN (Network in Network) architecture. The proposed Guided filter.
approach[1] yields 1.31-1.53x improvement in training energy
In this paper[4], they have shown a sub-window box filter
in comparison to conventional CNN implementation. They
whose weight function is independent from image content and
also obtain improvement up to 1.4x in training time, up to
its support regions are adaptive on image content. This filter is
2.23x in storage requirements, and up to 2.2x in memory
a little bit slower than box filter but can preserve corners and
access energy. The accuracy degradation suffered by the
edges during image smoothing. We analyze the role of its
approximate implementations is within 0-3% of the baseline.
parameters and compared it with guided filter on real images.
Handwriting recognition is a challenge task due to the large The sub-window box filter can be implemented on GPU if
variability in human writings. Improving the feature further performance is required. Thanks to its computation
representations that rely on the visual appearance of the efficiency and edge preserving property, this filter is suitable
handwritten text would lead to better recognition. Mohammed for a large range of image processing problems.
O. Assayony et al.[2],integrates two powerful appearance-
Box filters have been used to speed up many computation-
based features for producing robust statistics for handwritten
intensive operations in Image Processing and Computer
text. A handwritten text image is filtered by a set of Gabor
Vision. They have the advantage of being fast to compute, but
filters of different scales and orientations for extracting
their adoption has been hampered by the fact that they present
texture-based local features. The Gabor filter response features
serious restrictions to filter construction. This paper[5] relaxes
are organized into two layouts, viz. the Statistical Gabor
these restrictions by presenting a method for automatically
Features and Gabor Descriptors, and fed to the Bag-of-
approximating an arbitrary 2-D filter by a box filter. To
Features for learning robust statistical representations for the
develop the method proposed by Bernardo Rodrigues Pireso
handwritten text. The produced features are utilized in a
et al.[5] ,they first formulate the approximation as a
holistic handwritten word recognition system and evaluated on
minimization problem and show that it is possible to find a
a public dataset of Arabic handwritten subwords of Arabic
closed form solution to a subset of the parameters of the box
checks legal amounts. The best average recognition accuracy
filter. To solve for the remaining parameters of the
achieved by the produced features is 86.44% which is
approximation, they developed two algorithms: Exhaustive
promising in such challenge dataset of large number of classes.
Search Exhaustive Search algorithm, that is well suited for
Chandranath Adak et al.[3] introduces an efficient edge approximating small filters; and the Directed Search
detection method based on Gabor filter and rough clustering. algorithm, that is well suited to approximating large filters.
The input image is smoothed by Gabor function, and the Experimental results show the validity of the proposed
concept of rough clustering is used to focus on edge detection method.
with soft computational approach. Hysteresis thresholding is
Kunal Narayan Chaudhury et al.[6]demonstrate that it is
used to get the actual output, i.e. edges of the input image. To
possible to filter an image with an elliptic window of varying
show the effectiveness, the proposed technique is compared
size, elongation and orientation with a fixed computational
with some other edge detection methods.
cost per pixel. Their method involves the application of a
The proposed technique[3] is tested on different images (e.g. suitable global pre-integrator followed by a pointwise-
medical images: MRI, CR, CT, X-Ray; remotely sensed adaptive localization mesh. They present the basic theory for
images: islands, urban areas, country, planets, i.e. satellite the 1D case using a B-spline formalism and then appropriately
images; real life object images: flower, egg, wheel, color bars, extend it to 2D using radially-uniform box splines. The size
house, household things etc.). It produces stable, more and ellipticity of these radially-uniform box splines is
adaptively controlled. Moreover, they converge to Gaussians Jian Chen et al.[9] presents appropriate line segmentation for
as the order increases. Finally, they present a fast and practical intersections (X-junctions) and branches (T-junctions). In the
directional filtering algorithm that has the capability of local regions of inter- sections and branches, multiple
adapting to the local image features. orientations occur. A novel representation called "orientation
space" is proposed, which is derived by adding the orientation
Meng-Chun Lin etal.[7] addresses on VLSI design of rank- axis to the abscissa and the ordinate of the image. The
order filtering (ROF) with a maskable memory for real-time orientation space representation is constructed by treating the
speech and image processing applications. Based on a generic orientation parameter, to which Gabor filters can be tuned, as
bit-sliced ROF algorithm, the proposed design[7] uses a a continuous variable. The problem of multiple orientation
special-defined memory, called the dual-cellrandom-access line segmentation is dealt with by thresholding 3D images of
memory (DCRAM), to realize major operations of ROF: the orientation space and then detecting the connected
threshold decomposition and polarization. Using the memory components therein. In this way, X-junctions and T-junctions
oriented architecture, the proposed ROF processor can benefit are able to be separated effectively. Experimental results are
from high flexibility, low cost and high speed. The DCRAM presented using synthesized and real biomedical images. In
can performthe bit-sliced read, partial write, and pipelined particular, overlapping vessels in an x-ray coronary
processing. The bit-sliced read and partial write are driven by angiogram were well segmented by orientation space filtering.
maskable registers. With recursive execution of the bit-slicing
read and partial write, the DCRAM can effectively realize The bilateral filter is a nonlinear filter that smoothes a signal
ROF in terms of cost and speed. Theproposed design has been while preserving strong edges. It has demonstrated great
implemented using TSMC 0:18 mm 1P6M technology. As effectiveness for a variety of problems in computer vision and
shown in the result of physical implementation, the core size is computer graphics, and a fast version has been proposed.
356:1 427:7 mm2 and the VLSI implementation of ROF can Unfortunately, little is known about the accuracy of such
operate at 256MHz for 1.8V supply. acceleration. In this paper, Sylvain Paris et al.[10] propose a
new signal processing analysis of the bilateral filter, which
Eustace Painkrasetal.[8] presents a VLSI architecture for complements the recent studies that analyzed it as a PDE or as
computing the Gabor jets for facial feature points from video a robust statistics estimator. Importantly, this signal-
containing faces for computer vision applications. The processing perspective allows us to develop anovel bilateral
proposed architecture maps well onto a chain of highly filtering acceleration using a downsampling in space and
pipelineable Processing Elements (PE) based on multiply- intensity. This affords a principled expression of the accuracy
accumulation units with in-built ROMs. It is thus well suited in terms of bandwidth and sampling. The key to our analysis
for VLSI implementation, either in algorithm-specific ASIC or is to express the filter in a higher-dimensional space where the
reconfigurable FPGA for signal intensity is added to the original domain dimensions.
The bilateral filter can then be expressed as simple linear
Gabor filtering, which is one of the widespread feature
convolutions in this augmented space followed by two simple
extraction techniques for face processing applications such as
nonlinearities. This allows us to derive simple criteria for
face tracking and face recognition? Depending on the number
down sampling the key operations and to achieve important
of Facial Feature Points (FFP) at which the Gabor jets are to be
acceleration of the bilateral filter. They show that, for the
computed as well as the number of scales and orientations of
same running time, our method is significantly more accurate
the constituent Gabor filters, the number of PEs can be scaled
than previous acceleration techniques.
and matched to suit the processing requirements, without
sacrificing on the computational latency. Due to the regular
data flow and modular structure, the linear array of PEs III.PROPOSED SYSTEM
performing Gabor jet computations can achieve 100%
hardware utilization and provide sustained throughput to match 3.1 Objectives
the real-time video frame rate. This VLSI architecture is
 To design Gabor filter and Box Filter for Image
mainly targeted for incorporation into a hardware
Processing Applications.
implementation of Dynamic Face Tracking System (DFTS)  To compare these filters in terms of area, power and
based on Gabor wavelets. The computation of Gabor jets is a delay.
compute intensive task. This computational burden can be  To verify the implementation of various filters using
eased by adopting a VLSI hardware implementation, there by Spartan-3 xc3s400 FPGA Platform.
achieving real-time performance with moderate circuit
3.2 What is Gabor Filter?
complexity. In this paper[8], they have proposed a VLSI
Architecture for a novel Gabor jet module with a bank of 10 In image processing, a Gabor filter, named after Dennis
Gabor PEs, that forms the backbone of our 3-tier Dynamic Gabor, is a linear filter used for edge detection. Frequency and
Face Tracking System. The Gabor PE operating at 100 MHz orientation representations of Gabor filters are similar to those
based on Gabor wavelet transformation helps to compute the of the human visual system, and they have been found to be
Gabor jets at the rate of 2 ms/FFP which in turn, helps to particularly appropriate for texture representation and
achieve real-time face tracking at 70 fps in a video sequence. discrimination. In the spatial domain, a 2D Gabor filter is a
Gaussian kernel function modulated by a sinusoidal plane
wave. Simple cells in the visual cortex of mammalian brains
can be modeled by Gabor functions. Thus, image analysis
with Gabor filters is thought to be similar to perception in the
human visual system

3.3 Digital Gabor Filter

Digital Gabor Filter was designed by transforming the design

into verilog using xilinx. The target device is Spartan 3 family.
Basically there were 3 major parts in the filter: CLU, ALU and
MEMORY. The ‘convolution’ signal indicates the operation of
the filter. If the signal is high then the convolution process
takes place. If it is low then the filter receives image input and
stores it to the memory based on the input location. The data
enters the filter pixel by pixel. The ‘PIXEL_X’ and
‘PIXEL_Y’ signal gave the address of the memory location.

3.4 Block Diagram of Gabor Filter

Fig.3.2.Gabor Processing Element

3.5.1.3 Gabor Filter Masks:

The Gabor filter masks were first evaluated in Matlab to

obtain the exact floating point Gabor coefficients. These filter
masks (40 in number) have complex coefficients which are
normalized and truncated to fixed point numbers with a
precision of 12 bits in order to achieve the numerical
accuracy.

3.5.1.4 Gabor Filter ROM:

Fig 3.1. Block Diagram of the proposed Gabor Filter This forms an integral part of each PE. Each ROM stores 2
coefficients corresponding to the real and imaginary parts of 4
3.5.1 Block Diagram Description: 15 x 15 pixel Gabor filter masks. Though shown in Figure 3
as separate ROMs, both the real and imaginary parts of the
The following are the components of the Gabor Filter: filter coefficients will be stored in a single ROM for faster
read access.
3.5.1.1 Image RAM:
3.5.1.5 Input Image Pixel Access:
The 15 x 15 pixel image to be convolved with various filter
masks is stored in an internal image RAM before computation In our architecture, 10 PEs are deployed to compute 10 of
begins. All the computations are performed with 8 bit gray these Gabor filter decompositions in parallel, each utilizing
level pixel values. Image RAM needs to be updated with only the same pixel simultaneously. This ensures that each of the
one row/column of 15 pixels for subsequent Gabor jet 225 pixels is read only once for all the 10 PEs in a sequential
computation for the next pixel in the search area of that manner. For the next set of 10 filters, these input image pixels
particular FFP. are read once again, thereby the Gabor jet computation for 1
pixel, for all 40 filters requires 4 read cycles. The choice of10
3.5.1.2 Gabor PE: PEs limits the fanout of each pixel to 10.

Fig 3.2 shows the basic processing element of Gabor filter 3.5.1.6 Gabor Jet Memory Write:
bank for computation of Gabor jet. Each PE consists of (i) 2
multipliers for real and imaginary multiplications with the The 10 Gabor jet elements computed every 225 cycles are
incoming pixel intensity, (ii) two adders with output registers, written into the Gabor jet RAM for further processing by the
each of 12 bit width, to do the summation and accumulation similarity computation unit. For each pixel in the search area
and (iii) one filter ROM of size 21600 bits (225 x 4 locations, of a FFP, the Gabor jet RAM requires 40 (Gabor jet elements)
each of width 24 bits). Each PE computes and accumulates 225 x 12 (bits/jet)x 2 (real and imaginary). For n x n search area
intermediate values and then outputs the Gabor coefficient for around a FFP, we compute n x n Gabor jets. This RAM can be
1 pixel and 1 Gabor filter at the end of 225 clock cycles before optimized when coupled with the pipelined similarity
calculating the Gabor coefficient of the same pixel for the next computation unit.
Gabor filter.
3.5.1.7 Input and Output Sequencing:

A Finite State Machine (FSM) controller takes care of the

proper input and output sequencing, as well as the timely
collection of results. The 225 pixels corresponding to a FFP
are read from the frame buffer and stored in the image RAM.
The Gabor jet module processes this data for 10 Gabor filter
operations and their output is written to the Gabor jet RAM at The proposed system is implemented using Xilinx.After
the end of every 225 clock cycles. In this manner, all the 40 implementing the above block diagram discussed for gabor
filters are evaluated using the same set of pixel values in a total filter the simulation results obtained is shown in the Fig.4.1
of 900 clocks. For the next pixel in the search area of the FFP,
below.
only n pixels in the Image RAM need to be updated and the
computation of the next Gabor jet takes place, until all the n x
n pixels are calculated. Subsequently, the module computes in
the above-mentioned method, the Gabor jets for the next FFP
and writes to the Gabor jet RAM, for consumption by the
similarity computation unit.

3.6 Box Filter

Box filtering is basically an average-of-surrounding-pixel kind

of image filtering. It is actually a convolution filter which is a
commonly used mathematical operation for image filtering. A
convolution filters provide a method of multiplying two arrays
to produce a third one. In box filtering, image sample and the
filter kernel are multiplied to get the filtering result. The filter Fig.4.1 Simulation Results of Gabor Filter
kernel is like a description of how the filtering is going to
happen, it actually defines the type of filtering. The power of The synthesis report about the timing summary and timing
box filtering is one can write a general image filter that can do details that are obtained are shown in the Fig.4.2.
sharpen, emboss, edge-detect, smooth, motion-blur, etcetera.
Provided appropriate filter kernel is used.

Where temporal filtering uses multiple versions of each pixel

value at the same position but different “times”, spatial
filtering uses pixel at the same “time” but at different spatial
coordinates in the image. When temporally filtering a static
scene, pixels from the same position but different times are
expected to have same noiseless value. When only a single
noisy image is available, spatial filtering is used. The simplest
spatial filter is the averaging filter which replaces a pixel with
the average of itself and the pixels in the local neighborhood.
The most common averaging filter is the box filter, which
gives equal weight to the 9 pixels in a 3*3 window. This
choice of weights maximally reduces the noise variance when
the noise in additive. Box filtering is similar to temporal
filtering except that samples from adjacent pixel positions are
used as approximations to multiple samples of the pixel at the Fig 4.2 Synthesis Report of Gabor Filter
same position.

3.6.1 Advantages of Box Filter:

Design overview summary of logic utilization and logic
 Exclusively designed for FPGAs with distributed and distribution of Gabor Filter is shown in the Fig 4.3 below.
block RAMs
 Less memory usage
 High processing speed
 Low computational complexity.

3.7 Applications of Digital Filters:

Digital image processing technology is evolving process in all

fields like
 Fingerprint authentication system
 Face recognition
 Iris recognition
 Speaker recognition
 Digital watermarking
 Medical image processing
 Industrial machine vision applications
 Application of image processing in remote sensing
 Image processing for remotely sensed data
 Artificial neural networks in image processing
Fig 4.3Summary on Device Utilizationof Gabor Filter

IV.RESULTS AND DISCUSSIONS The power report obtained on Gabor is shown in the Fig 4.4
below.
 Fig 4.4 Power Report on Gabor Filter

After implementing the above block diagram discussed for box Fig 4.7 Summary on Device Utilization of Box Filter
filter the simulation results obtained is shown in the Fig.4.5
below. The power report obtained on Box Filter is shown in the fig
4.8 below.

Fig.4.5 Simulation Results of Box Filter

The synthesis report about the timing summary and timing

details that are obtained are shown in the Fig.4.6. Fig 4.8 Power Report on Box Filter

Table I shows the comparative study of area, power and delay

of both Gradient Filter and Box Filter.

TABLE I: Comparison of Resources of Gradient Filter and

Box Filter

Area Power Delay

Gabor LUTs-279 391mW 13.414ns
Filter
Box LUTs-460 178mW 4.855ns
Filter
It can be observed from the Table I that Box filter performs
well in terms of Power and delay compared to Gabor Filter.

Fig 4.9 and Fig.4.10 below shows the example of an image

Fig 4.6 Synthesis Report of Box Filter
obtained using Gabor and Box Filter in Matlab.
Design overview summary of logic utilization and logic
distribution of Box Filter is shown in the Fig 4.7 below.
 REFERENCES

[1]. Syed ShakibSarwar, Priyadarshini Panda and

KaushikRoy,"Gabor Filter Assisted Energy Efficient Fast
Learning Convolutional Neural Networks",EEE/ACM
International Symposium on Low Power Electronics and
Design (ISLPED), Taipei, 2017, pp. 1-6.

[2] Mohammed O. Assayony, Sabri A. Mahmoud,Integration

of Gabor Features with Bag-of-Features Framework for
Arabic Handwritten Word Recognition,9th IEEE-GCC
Conference and Exhibition (GCCCE) ,2017 .

[3] ChandranathAdak,"Gabor Filter and Rough Clustering

Based Edge Detection",International Conference on Human
Fig 4.9. Example–Gabor Filter using Matlab Computer Interactions (ICHCI), 2013.

[4] Yuanhao Gong, Bozhi Liu, XianxuHou,

GuopingQiu,"Sub-window Box Filter", IEEE Visual
Communications and Image Processing (VCIP),2018.

[5] Bernardo Rodrigues Pires, Karanhaar Singh and Jos´e M.

F. Moura,"APPROXIMATING IMAGE FILTERSWITH
BOX FILTERS",18th IEEE International Conference on
Image Processing,2011.
[6] Kunal Narayan Chaudhury, Arrate Munoz Barrutia and
Michael Unser,"FAST ADAPTIVE ELLIPTICAL
FILTERING USING BOX SPLINES",15th IEEE
International Conference on Image Processing,2008 .

[7] Meng-Chun Lin, Lan-RongDung,"On VLSI design of

Fig 4.10 Example – Box Filter using Matlab rank-order filtering using DCRAM
architecture",Elsevier,INTEGRATION, the VLSI journal 41
It can be observed that the quality of the output image after (2008) 193–209DOI: 10.1016/j.vlsi.2007.05.002
box filtering is good compared to Gabor filter. The Gabor filter
[8] EustacePainkras and CharayaphanCharoensak,"A VLSI
suffers with poor image quality after filtering. Architecture for Gabor Filtering in Face Processing
Applications", International Symposium on Intelligent Signal
VIII. CONCLUSION Processing and Communication Systems, 2005.
According to current technology future is totally based on [9] JianChen,YoshinobuSato,ShinichiTamura,"Orientation
virtual world. Right now everything is based on real data Space Filtering for Multiple Orientation Line
transfer and as we know for real time data transfer we need Segmentation",IEEE Transactions on Pattern Analysis and
noise less system. Many image filtering can be effectively Machine Intelligence • June 2000.
implemented with a reduced number of operations per pixel. In
this paper we have done the comparison between image [10] Sylvain Paris and FredoDurand,"A Fast Approximation
filtering. The enhanced design proposed for Gabor Filter and of the Bilateral Filter using a Signal Processing Approach",
Box Filter has implemented. The processing of images is faster European Conference on Computer Vision, ECCV 2006, pp
and more cost-effective. The area of the design has been 568-580.
significantly reduced while the function of the filter is
perfectly maintained. According to this paper we did a
comparative study about the gabor and box filters which is
time complexity with maintain the quality level of the
generated output images. By comparing various filtering
techniques it can be concluded that box filter is better than
gabor filter. It will make a better output by smoothening and
sharpening the image in an efficient manner and also it
diminishes noises. The same has been successfully verified in
Spartan-3 FPGA Kit.

Future Scope

Even though box filters produce good image quality, steps

must be taken to reduce the LUT size in the Box Filter. In
future, box filter can further be enhanced to be used along with
other digital image filters namely gradient and guided Image
filter in order to achieve reduced area.