ISSN (Print) : 2320 – 3765

ISSN (Online): 2278 – 8875

International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering

Vol. 2, Issue 5, May 2013

Triple Input Sorter Optimization

Algorithm of Median Filter Based on
FPGA
Chandana. Raj1, Dr. Keshaveni. N2
PG Student [VLSI Design], Dept. of ECE, East point college of Engineering, Bangalore, Karnataka, India 1
Professor, Dept. of ECE, East point College of Engineering, Bangalore, Karnataka, India 2

ABSTRACT: This project is focused on developing hardware implementations of image processing algorithm for use
in an FPGA -based image processing system, this approach facilitates comparison of the software and synthesized
hardware algorithm outputs. The rank order filter is a particularly common algorithm in image processing systems. This
approach was taken because it speeds understanding of the algorithm design, reduces the number of comparison, and
the area required. For every pixel in an image, the window of neighbouring pixels is found. Then the pixel values are
sorted in ascending, or rank, order. Next, the pixel in the output image corresponding to the origin pixel in the input
image is replaced with the value specified by the filter order. The main advantage of the sorting network is that the
sequence of comparisons is fixed. Thus it is suitable for parallel processing and hardware implementation, especially if
the number of sorted elements is small. The number of compare–swap components and delay are two crucial
parameters of any sorting network. By delay we mean the minimal number of groups of compare–swap components
that must be executed sequentially. This algorithm ensures the number of comparators and delay used for given input
and reduces the number of comparisons and computation time.

Keywords: spatial filtering, window, 2D mask, kernel, impulsive noise.

I.INTRODUCTION
Although seemingly easy for living beings, coping with vision is less natural for computers. Images need to be
captured, digitized and processed until some form of decision or conclusion can be made. Thus image processing, the
analysis and manipulation of data, originally in the form of an image or image sequence, by a computer is
acknowledged as one of the grand challenges of computing. The reasons for this are: Data size, Computation
complexity, Time constraints, Types of operations, Variety in image processing algorithms, Different data types. Image
processing tasks such as filtering, stereo correspondence and feature detection are inherently highly parallelisable. The
use of FPGAs (Field Programmable Gate Arrays), which can be operated in highly parallel configurations, can thus be
a useful approach in imaging applications. The median filter is a non-linear filter; it is a special case of rank order
filters whose rank is half the length of the sequence. In image processing applications, median filter is used to remove
impulsive noise from images while preserving the edges. One of the disadvantages of linear filters, such as the moving
average filter, when used to denoise the data, is that they not only smooth the noise, but also smooth the sudden and
sharp transitions that were present in the original data, such as edges in images. Moreover, they are not as efficient as
the median filters in removing certain types of noise, such as impulsive noise Although median filters do not blur the
edges as much as the linear filters do, because they still possess smoothing characteristics, as the size of the filter
increases, there may be significant image blurring. In image processing, the 2D filtering operation is performed by
sliding the window along all the rows and columns of the image until all the pixels are covered by the window. Median
filtering techniques are usually based on the sorting network architectures; another approach to implement the median
filter is based on the histogram. The resources required to implement the sorting network architecture on a FPGA
device increases with the size of the filtering window. However, the sorting network based algorithms are independent
of the size of the image and depend only on the size of the window.

II.DESCRIPTION OF MEDIAN FILTER

Median filter is often applied to gray value images due to its property of edge preserving smoothing. The main idea of
the median filter is to run through the signal entry by entry, replacing each entry with the median of neighbouring
entries. The pattern of neighbours is called the "window", which slides, entry by entry, over the entire signal. For 1D
signals:, the most obvious window is just the first few preceding and following entries, for 2D (or higher-dimensional)
signals such as images, more complex window patterns are possible (such as "box" or "cross" patterns). If the window
has an odd number of entries, then the median is simple to define: it is just the middle value after all the entries in the
Copyright to IJAREEIE www.ijareeie.com 1970
 ISSN (Print) : 2320 – 3765
ISSN (Online): 2278 – 8875

International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering

Vol. 2, Issue 5, May 2013

window are sorted numerically [2]. For an even number of entries, there is more than one possible median,
see median for more details. Like the mean filter, the median filter considers each pixel in the image in turn and looks
at its nearby neighbours to decide whether or not it is representative of its surroundings. Instead of simply replacing the
pixel value with the mean of neighbouring pixel values, it replaces it with the median of those values [1]. The median is
a more robust average than the mean and so a single very unrepresentative pixel in a neighbourhood will not affect the
median value significantly. Since the median value must actually be the value of one of the pixels in the
neighbourhood, the median filter does not create new unrealistic pixel values when the filter straddles an edge. For this
reason the median filter is much better at preserving sharp edges than the mean filter. In general, the median filter
allows a great deal of high spatial frequency detail to pass while remaining very effective at removing noise on images
where less than half of the pixels in a smoothing neighbourhood have been effected. To find the median it is necessary
to sort all the values in the neighbourhood into numerical order and this is relatively slow, even with fast sorting
algorithms such as quick sort. The basic algorithm can, however, be enhanced somewhat for speed. A common
technique is to notice that when the neighbourhood window is slid across the image, many of the pixels in the window
are the same from one step to the next, and the relative ordering of these with each other will obviously not have
changed. To apply the mask means to centre it in a pixel, evaluating the covered pixel brightness’s and determining
which brightness value is the median value [4]. The median value is determined by placing the brightness’s in
ascending order and selecting the centre value.

III.ARCHITECTURE
Requiring on the median value, then there is no need for sorting the .whole window. An optimized version of the
Bubble sorter is presented in Figure 1, for the special case of a 3x3 window. The architecture is based on a Triple Input
Sorter [6] and consists of the following stages Stage 1: Sorts the elements of each window column. Stage 2: Sorts the
elements of each window row. Stage 3: The median is finally obtained by sorting the cross diagonal and picking up the
middle value. Only three results are guaranteed to be placed in the right place: the maximum, the minimum and the
median; this saves a great deal of logic. The main component in this design is a Triple Input Sorter (which is based on a
dual input sorter), Triple Input Sorter based algorithm (TIS) which is an optimized version of the Bubble sort algorithm
for the special case of a 3x3 window size which is as shown in fig2. During the median filter step 9 neighbouring
pixels including the centre pixel are assigned to three row extractors for shortening the searching time of the median
value. At first, each row extractor extracts the median value of three pixels in its row. The three row extractors work in
parallel, which is one operation level parallel. Then, the final median extractor calculates the median value of the output
values of three row extractors. This improvement will greatly expedite the process of median searching [5].

Fig. 1 Median filter window with pixels arranged according to their intensities

Copyright to IJAREEIE www.ijareeie.com 1971

 ISSN (Print) : 2320 – 3765
ISSN (Online): 2278 – 8875

International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering

Vol. 2, Issue 5, May 2013

Fig. 2 block diagram of 3x3 triple input sorter based median filter design
Equations:
In simple terms , scanning through the image pixels by a 3 * 3 matrix calculate the median by sorting the values then we
replace the pixel being considered with the middle pixel value (if we have even number of pixels then we use average of
the two middle pixel), as shown in fig1

The classic median filter is also a subset of a more general set of filters, called rank order filters, which are themselves
a subset of stack filters [7]. As in median filtering, a 2-D rank order filter consists in passing a PxQ window over an
image and computing a result pixel from the sample pixel values S = (S1, S2, ……, SP*Q) within the window.
The basic 2D (3 * 3) median filter is characterized by the following equation:
OUTPUT(x, y) = MED{ PI, P2, P3,
P4, P5, P6,
P7, P8, P9)
OUTPUT(x, y), which is the median data of these nine pixels, will replace the central point P5.
OUTPUT(x, y) = MED { MIN[S], P5 , MAX[S] }
Where S is the set of samples surrounding the central point except its horizontal neighbors (P4, P6),
S = {PI, P2, P3, P7, P8, P9)
MIN[S] is the minimum data of S, and MAX[S] is the maximum data of S.

IV.SYSTEM IMPLEMENTATION
A. System overview: Prior to any hardware design, the software versions of the algorithms are created in MATLAB.
Using MATLAB procedural routines to operate on images represented as matrix data, these software algorithms
were designed to resemble the hardware algorithms as closely as possible. This approach was taken because it
speeds understanding of the algorithm design. In addition, this approach facilitates comparison of the software and
synthesized hardware algorithm outputs. When the median filter must be carried out in real time, the software
implementation in general purpose processors does not usually give good results.

Copyright to IJAREEIE www.ijareeie.com 1972

 ISSN (Print) : 2320 – 3765
ISSN (Online): 2278 – 8875

International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering

Vol. 2, Issue 5, May 2013

Fig. 3 Block diagram of Image median filtering

Reconfigurable computing architectures [1], [2] are sufficiently flexible so that new operations can be implemented in
the existent hardware, and they are quite quick for real-time execution. Conversion of a colour image to gray scale is
not unique; different weighting of the colour channels effectively represent the effect of shooting black-and-white film
with different-colored photographic filters. Real time images from web cam are taken and processed in matlab which
first converts the real time color image to gray scale, which is in turn converted to text file which is given as a input to
median filter on FPGA through Xilinx which process in finding the median value of image by removing the salt and
pepper noise present in image without any blurring of image by edge preservation and is back converted using matlab.

B. Hardware implementation of median filter module: A sorting algorithm is an algorithm that puts elements of a list in
a certain order. Efficient sorting is important for optimizing the use of other algorithms (such
as search and merge algorithms) which require input data to be in sorted lists. Bubble sort is a simple sorting
algorithm. Median filtering techniques are usually based on the sorting network architectures, These sorting networks
use compare and delay units to implement the median filter. The incoming pixels are passed through a network of
comparators and swapping units - the comparator units compare two to three incoming pixels at once and then the
swapping unit sorts them accordingly. The algorithm starts at the beginning of the data set. It compares the first two
elements, and if the first is greater than the second, it swaps them. It continues doing this for each pair of adjacent
elements to the end of the data set. It then starts again with the first two elements, repeating until no swaps have
occurred on the last pass. Because we can get the maxim value or the minimum value of 3 data by doing 2
comparisons, and get the median value or the ordering of 3 data by doing 3 comparisons, we design 4 different
comparators, such as ordering comparator, minimum comparator, median comparator and maxim comparator [5].
The hardware structures of the 4 comparators are given in figure 3. the processing element is basic processing unit,
which is designed for comparing 2 input data. D stands for D flip-flop, whose function is to make a single-circle
delay. It is used for synchronizing the calculation here. Because the proposed algorithm needs to storage two groups
temporary data in current processing, we design two D flip-flops to realize this function [1]. The data’s will be sent to
the ordering comparator ordinally for data ordering, and the results will be send to next different comparators.
Before the second comparing, the data need 2 circle delay to distinguish the circle order of input data by two D flip-
flops. The second comparing results will be sent to the final median comparator to get the final result.

Copyright to IJAREEIE www.ijareeie.com 1973

 ISSN (Print) : 2320 – 3765
ISSN (Online): 2278 – 8875

International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering

Vol. 2, Issue 5, May 2013

Fig. 4 Hardware module of median filter

The median value will be the middle value of the sorting network. It utilize 3×3 masks, which means that processing
each pixel, as the pixels in the boundary of each section, needs eight pixels around it. In order to deal with the boundary
problem, each section contains four more rows of pixels than half input image. The four extra rows are responsible for
processing a boundary row.

V.RESULT
This algorithm is applied for 3x3 median filter on a real time image. since median filter Is very effective in removing
the salt and pepper noise more effectively in gray scale the real time color image is converted to gray scale and the
processing is carried out as shown in fig. 5.a

Fig. 5.a. color to grayscale conversion

Copyright to IJAREEIE www.ijareeie.com 1974

 ISSN (Print) : 2320 – 3765
ISSN (Online): 2278 – 8875

International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering

Vol. 2, Issue 5, May 2013

Fig 5. b Text format of grayscale image

The fig.5.b shows the text format of the gray scale image done in Matlab, which is then given as input to
FPGA for median filtering.

Fig. 5. c simulation results of grayscale image

Fig.5.c shows the simulation results of the median value calculation with very less computation from the first output to
the last output, which has the great impact on the speed of image processing on real time application.
The computation time from reset output to the first output available is comparatively smaller than the traditional median
filtering algorithm.

Fig. 5.d. Matlab Results of median filter with filtered and denoised output image in grayscale
Fig 5.d shows the Matlab output with real time color image getting converted to grayscale and median filtering on FPGA
is performed, as it is very effective in removing the impulsive noise like salt and pepper noise without any blurring of
original image.
Copyright to IJAREEIE www.ijareeie.com 1975
 ISSN (Print) : 2320 – 3765
ISSN (Online): 2278 – 8875

International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering

Vol. 2, Issue 5, May 2013

VI.CONCLUSION
The experimented results show that it can effectively reduce the issues of real time applications, with less computation
time by reducing the number of unnecessary comparisons, pixel memory and the size of the memory which are made
independent of any image sizes. The image median filtering is done in both on FPGA and Matlab which provides a
door to comparison of hardware and software results. Despite the effectiveness of spatial domain technique s when applied separately,
in practice a combination of such methods can be used to achieve more effective image enhancement. This effectively removes the
impulsive noise and preserving the edges with smaller computation time and lesser area required. This approach is
highly parallelizable. Nonlinear filters offers a flexible, robust approach to the problem of estimating signals in the
presence of impulsive noise.

ACKNOWLEDGMENT
This paper is made possible through the help and support from parents, teachers, friends I sincerely wish to thank
them all.

REFERENCES
[1] P. WEI, L. ZHANG, C. MA, AND T. S. YEO, “FAST MEDIAN FILTERING ALGORITHM BASED ON FPGA,” IN SIGNAL PROCESSING (ICSP), 2010
TH
IEEE 10 INTERNATIONAL CONFERENCE , PP. 426–429, 2010.
[2] C. J. Juan, “Modified 2d median filter for impulse noise suppression in a real-time system,” Consumer Electronics, IEEE Transactions on,
vol. 41, no. 1, pp. 73–80, Feb. 1995.
[3] R. Maheshwari, S. Rao, and P. Poonacha, “Fpga implementation of median filter,” in VLSI Design, Tenth International Conference 1997,
pp. 523–524, Jan. 1997.
[4] Miguel A. Vega-Rodríguez, Juan M. Sánchez-Pérez, Juan A. Gómez-Pulido “An fpga-based implementation for median Filter meeting
the real-time requirements of Automated visual inspection systems” , Proceedings of the 10th Mediterranean Conference on Control and
Automation – MED2002 Lisbon, Portugal, July 9-12, 2002
[5] Y. Lu, M. Dai, L. Jiang, and S. Li, “Sort optimization algorithm of median filtering based on fpga,” in Machine Vision and Human-
Machine Interface (MVHI), 2010 International Conference, pp. 250– 253, 2010.
[6] K Benkrid, D Crookes School of Computer Science, The Queen’s University of Belfast, “A New Bit-Level Algorithm for General
Purpose Median Filtering” Belfast BT7 1NN, UK.
[7] K. R. Castleman, “Digital Image processing”, Prentice Hall, ISBN: 0132114674, 1995.

Copyright to IJAREEIE www.ijareeie.com 1976