International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 4 - November 2015 A review on fuzzy image compression Linsa v u#1, Dr.T.M Amarunnishad*2 # M.Tech T.K.M College of Engineering Kollam, Kerala, India Abstract — Image compression is a technique to reduce the data of representation of images. Different methods are available for compression of images. Digital image compression used in number of applications, like satellite communication, huge storage and transformation. The good compression method should be fast and memory efficient. It also reduces storage and transmission space and reduces transmission bandwidth. The fuzzy based image compression consists three steps .fuzzification; modify the pixel value and defuzzification. In compression the modification means reduce the number of bits to represent the pixel or reduce the number of pixels. In this study discuss various fuzzy image compression techniques that reduce the size of image without affect the visual quality of image. Keywords — fuzzy image processing ; fuzzy image compression;fuzzy relational equations; fuzzy transform; vector quantization, . I. INTRODUCTION Digital image compression are useful in many field such as satellite images, medical images ,huge data transfer and sharing ,storage etc .Hence we need efficient very fast and memory efficient(highly compressed) image compression. Image compression reduces the data of representation of images. it reduces the irrelevant data in the image .Image compression reduces [1]either inter pixel redundancy psycho visual redundancy coding redundancy Basically image compression is classified into lossy and lossless [1]. Lossy image compression cannot reconstruct the exact copy of the original image. This reconstructs the image with loss of some information compare to the original one. so lossy image compression is suitable in exact information is not needed applications. Important lossy image compression methods are transform coding, vector quantization ,block truncation coding etc. compression methods are transform coding, vector quantization ,block truncation coding etc. But in lossless image compression that produce or decode the exact copy of the original image. That cannot loss the information of the input image. Lossless image compression is But in lossless image compression that produce or decode the exact copy of ISSN: 2231-5381 the original image. That cannot loss the information of the input image. Lossless image compression is useful in medical image processing like applications .example of lossless image compression methods are Huffman coding, arithmetic coding, run length coding etc. II. COMPONENTS OF IMAGE COMPRESSION The componets of image compression are encoder and decoder.The basic block diagram of compression shows in Fig.1.Fig.2 shows the reverse of compression (reconstuction) processs.The encode perform the compression operations and decoder perform the decompression process. Input image Compressi on Compresse d image Fig.1. Block diagram of compression Compress ed image Decompr ession Reconstruct ed image Fig.2. Block diagram of decompression III.FUZZY IMAGE COMPRESSION Fuzzy image processing means the fuzzy logic is applied to the image processing for better processing. it is a collection of different fuzzy techniques applied to the image processing. All the approaches understand, represent and process the images; the features are the fuzzy sets. The representation and processing depends the image processing applications. Basically fuzzy image processing has 3 steps .Fuzzification, modify the membership function and defuzzification. Fig.3 shows the basic block diagram of fuzzy image processing. Fuzzification is the change the pixel values for processing in fuzzy logic. The main power of fuzzy image processing depends the second step, modify the membership function. The modification is the actual process in http://www.ijettjournal.org Page 204 International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 4 - November 2015 image processing. Last one is defuzzification. It converts into the original form. Image Intermediate level Segmentation Representation Description Low level Preprocessing Geometrical fuzziness Grayness ambiguity compression we try to reduce the redundancy like inter pixel, psycho visual, coding redundancy. We reduce the number pixel in the second step or reduce number bits to represent the pixel in second step. Fuzzificati on Modificat ion Defuzzifi cation Fig.3. Fuzzy image processing IV. MEASURES Uncertainty Fig.4. block diagram of uncertainty There are so many reasons why we should use fuzzy image processing .the important reasons are It is a powerful tool for knowledge representation and processing. It manages efficiently. uncertainty and vagueness Fuzzy logic and fuzzy set theory provides a better method for image processing using if the rules. Other one is many difficulties occurs due to data /task/results are uncertain. Basically we can distinguish three kinds of imperfections in image processing. grayness geometrical vagueness High level Analysis Interpretation Recognition Result compression ratio time of processing memory need Compression ratio indicates how much compress particular image. For example if the compression rate 0.111 indicates the compressed image is 0.111 of the original image. The compression ratio is calculated by compressed image size/original image size. Computation time is the time required to transmit the original image into the coded image .that depends the different techniques. Objective quality of the image coding method is measured by following parametric. Memory requirement is the storage space need to store the image. Normally used measures for measuring objective quality of image in image compression are MSE PEN PSNR MSE(mean square error) is the cumulative squared error between the reconstructed(decoded) image and the original image .if the MSE less that indicate the good compression result .Equation(1) is the standard equation for calculate the MSE. Vague knowledge Fig.4 represents the block diagram of uncertainty in fuzzy image processing In fuzzy image compression, fuzzified by using any Fuzzification method that is applicable all fuzzy image processing. Only the difference in the main step modification of the membership functions. In fuzzy ISSN: 2231-5381 All title and author details must be in single-column format and must be centered. Different performance metrics are selected for measuring the objective and subjective quality of the coding/decoding system. The suitable method is selected based on PSNR (peak signal to noise ratio) is the ratio of signal to noise .it is calculated based on MSE.PSNR is high that indicate the good compression method. Equation for determine the PSNR is (2). PEN (penalty function) is the penalty function. Equation (3) specifies the equation for PEN. it is useful in color image processing. http://www.ijettjournal.org Page 205 International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 4 - November 2015 N i 1 cR,G ,B MSE M j 1 P (i, j) P E c NM c (i, j) frame (predictive frame).I frame is the original frame of motion .P- frame has slight difference from the Iframe. Similarity between P-frame and I-frame is calculated by equation(6) .if it has greater than a particular value the two frames are different-frame is compressed higher rate than the I-frame.the frame are compressed same like image compression . 2 3MN (1) iN1 Mj1 cR,G ,B Pc (i, j ) PcENM (i, j) PEN 2 3MN SimL ( F , G ) 1 m n 1 maxFij , Gij min Fij , Gij ( m n) i 1 j 1 (2) (6) max x , y I ( x, y ) PSNR 10 * log10 RMSE 2 (3) V. DIFFERENT FUZZY IMAGE COMPRESSION TECHNIQUES A. Fuzzy relation equation for compression In fuzzy relation equations based compression the Fuzzification of input pixel values same as fuzzy transform based compression. The input (source) image pixels are fuzzified by Original pixel value divided by 255, 255 is the maximum gray level value the fuzzified values are compressed by Equations of fuzzy relation (4). M N C pq A pi tBqj t S ij i 1 j 1 (4) The table 1 shows the image compression by fuzzy relational equations. The compressed image are reconstructed by Equation of fuzzy relation (5). K H Dij A pi tB qj tG pq p 1 q 1 (5) Basically the Images are compressed by fuzzy relational equation. From the equation of compression we can see different type fuzzy relational equation for compression is possible depends the norms. In [3] fuzzy relational equation is selected for compression of gray scale images. After that this method is extend to color images. In color images the pixels are transformed into the YUV space (RGB to YUV), using equations .then apply the fuzzy relation equation in each domain separately as same as the in [5]. In [22] fuzzy relational equations for coding and decoding process of images and videos. The gray scale images are coded by fuzzy relation equations like in [12].Using Gödel, goguen ,lukasiewicz t-norm. The case RGB images each component are separated then apply the fuzzy relation equation like gray images. Then combine the components. In video compression frames are separated as I-frame (intra frame) and P- ISSN: 2231-5381 In [23] find the efficient decomposition method of fuzzy relation equations .one optimization method improve the cost function. Another one improves the decomposition process. The PSNR value of this method (fuzzy relation equation) is compare with the JPEG and MPEG. B. Fuzzy transform based compression Fuzzy transform based compression is better than some other existing methods. It is lossy image compression method. so the reconstructed image is not a exact copy of the original image. In fuzzy transform based compression fuzzy transform of two variables is selected for compression .in this method the input values are image pixels or intensity values. The input pixel values are fuzzified in the first step .In[1,2]the pixels are divided by 255(maximum gray level value. the fuzzified value )for Fuzzification. Equation (7) denotes the fuzzy transform of two variables. The two input parameter are membership functions (normally use standard member ship functions sinusoidal, triangular membership functions) that represent in (8),(9),(10) .All[233] this techniques the selected membership functions are (8).The two variable fuzzy transform is applied in the output of Fuzzification .then get the compressed representation. Ckl Mj1 iN1 R(i, j ) Ak (i) B l ( j ) Mj1 iN1 Ak (i) B l ( j ) (7) 0.5(1 cos ( x x1 )) ifx x1 , x2 A1 ( x) h 0 otherwise (8) http://www.ijettjournal.org Page 206 International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 4 - November 2015 equation(6).In[12]segmentation process is used before compression. FGFCM (fast generalized fuzzy c-mean) clustering is used. The compressed by fuzzy transform. 0.5(1 cos ( x xk )) ifx xk 1 , xk 1 Ak ( x) h C. Fuzzy vector quantization for image compression 0 otherwise (9) 0.5(1 cos ( x x n )) ifx x n1 , x n An ( x) h 0 otherwise (10) The reduction of color images by fuzzy transform [7] .In image reduction the reduced (compressed) images via image compression techniques compare with the original image .but in image compression the reconstructed (decompressed) image is compare with the original image. The image are compressed by f-transform are magnified for comparing with the source or initial image. The compressed images are reconstructed by using inverse discrete F-transform for comparing with the original one. After comparison the efficiency (subjective and objective) of the techniques are determined. Equation (11) is the reconstruction equation in fuzzy transform. n m Dnm (i, j ) C kl Ak (i ) Bl ( j ) k 1 (11) l 1 The discrete F-transform is selected for compression in[9] and inverse fuzzy transform is selected for compression in[10].it is similar to the fuzzy transform based compression only the slight variation in equation. The objective measures are comparing with the JPEG method. Vector quantization is a type of lossy image compression. Normally it includes four steps. Vector formation, training vector selection, codebook generation and quantization. The input image pixels are divided and generate the vectors. From the set of vectors some vectors are selected for training purpose. Codebooks of codeword’s are generated based on standard clustering algorithms. Then the input vectors are selected and corresponding codeword selected form the codebook. Likewise reduce the total storage and transmission space. In [21] generate codebook using fuzzy c-mean clustering operates image histogram, is used to generate good codebook. In [16] improve the fuzzy cmean clustering. It reduces certain problems. it change the objective function of fuzzy c-mean clustering. Then change the overall training procedure. In [18] generate fast vector quantization procedures. For that reduce the number of codebooks that are affected by specific pattern. Also reduce the number of patterns in the design process. Sequential implementation above two method reduce the computational cost. In [17] combine the fuzzy clustering with competitive agglomeration and novel codebook migration strategy. It reduce the when using the fuzzy clustering method. In [19] use FPSO (fuzzy particle swarm optimization, itself extract the near optimum codebook of vector quantization. D. Other methods Wavelet transform coding is applicable for image compression. In [23] embedded zero wavelets coding is selected for compression. Fuzzy clustering applied In [8] discuss the theory and application of fuzzy to quantizing the wavelet coefficients after coded by transform. It includes fuzzy transform of two variables, wavelet transform. one variable, discrete and continues fuzzy In [22] region oriented compression of color images transforms .In [14] fragile water marking tamper using fuzzy inference and fast merging. Splitting detection with images compressed by fuzzy transform. based on watershed transform and merging using The image are compressed by fuzzy transform then fuzzy color preserving rule based system and novel the compressed image used in watermarking process. one dimensional graph structure. In [13] color image compression with fuzzy transform in YUV space. The RGB images are converted into the YUV space similar to the previous method. Then the converted image compressed. In[15] fuzzy transform for compression and decompression of color video .separate I-frame and P-frame and compress each frame using the fuzzy transform Pframe is more compressed compare to I-frame.the frames are separated by ISSN: 2231-5381 VI. CONCLUSIONS Here we discuss about various fuzzy image compression techniques. Fuzzy relational equation based, fuzzy transform based and vector quantization based. Fuzzy transform based compression provides better results than fuzzy relational equation based compression. in image compression we focus the http://www.ijettjournal.org Page 207 International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 4 - November 2015 higher compression rate with high quality of the output image. REFERENCES [1] G Rafael C .Gonzalez, Richard E. Woods ”Digital image processing” ,third edition. [2] Tizhoosh,H.R. “Fuzzy Image Processing “ Springer, 1997, ISBN: 3-540-63137-2. [3] Hajime Nobuhara, Witold Pedrycz, Salvatore Sessa, Kaoru Hirota ,“A motion compression/reconsruction method based on max t-norm composite fuzzy relational equations”, Information Sciences, Volume 176, Issue 17, 3 September 2006, Pages 2526-2552. [4] Vincenzo Loia, Salvatore Sessa,”fuzzy relation equation for coding/decoding processes of images and videos”, Information Sciences, Volume 171, Issues 1–3, 4 March 2005, Pages 145172. [5] Ferdinando Di Martino, Vincenzo Loia, Salvatore Sessa,”fuzzy relation equation for compression/decompression processes of colour images in the RGB and YUV colour spaces”, Information Sciences, Volume 180, Issue 20, 15 October 2010, Pages 3914-3931. [6] Hajime Nobuhara, Kaoru Hirota, Salvatore Sessa, Witold Pedrycz,” efficient decomposition methods of fuzzy relation and their application to image decomposition”, Applied Soft Computing, Volume 5, Issue 4, July 2005, Pages 399-408. [7] Ferdinando Di Martino, Petr Hurtik, Irina Perfilieva, Salvatore Sessa,” A color image reduction based on fuzzy transforms”, Information Sciences, Volume 266, 10 May 2014, Pages 101111. [8] Irina Perfilieva,” fuzzy transforms :theory and applications”, Fuzzy Sets and Systems, Volume 157, Issue 8, 16 April 2006, Pages 993-1023. [9] Ferdinando Di Martino, Salvatore Sessa, “ compression and decompression of images with discrete fuzzy transforms”, Information Sciences, Volume 177, Issue 11, 1 June 2007, Pages 2349-2362. [10] Ferdinando Di Martino, Vincenzo Loia, Irina Perfilieva, Salvatore Sessa ,” An image coding/decoding method based on direct and inverse fuzzy transforms”, International Journal of Approximate Reasoning, Volume 48, Issue 1, April 2008, Pages 110-131. [11] Ferdinando Di Martino, Salvatore Sessa ,” Fragile watermarking tamper detection with images compressed by fuzzy transform”, information Sciences, Volume 195, 15 July 2012, Pages 62-90. [12] Ferdinando Di Martino, Salvatore Sessa, “A segmentation method for images compressed by fuzzy transforms” information Sciences, Volume 195, 15 July 2012, Pages 62-90 ISSN: 2231-5381 [13] F. di martino, v. loia, s. sessa, “Direct and inverse fuzzy transforms for coding/decoding color images in YUV space”,journalof uncertain systems,Volume 3,no.1,pp.1130,2009. [14] Irina Perfilieva, Bernard De Baets, “ fuzzy transforms of monotone functions with application to image compression”, Information Sciences, Volume 180, Issue 17, 1 September 2010, Pages 3304-3315. [15] Ferdinando Di Martino, Vincenzo Loia, Salvatore Sessa,” fuzzy transforms for compession and decompression of color videos”, information Sciences, Volume 180, Issue 20, 15 October 2010, Pages 3914-3931. [16] George E. Tsekouras, Mamalis Antonios, Christos Anagnostopoulos, Damianos Gavalas, Dafne Economou, “improved batch fuzzy learning vector quantization for image compression”, information Sciences, Volume 178, Issue 20, 15 October 2008, Pages 3895-3907. [17] Dimitrios Tsolakis, George E. Tsekouras, John Tsimikas,” fuzzy vector quantization for image compression based on competitive agglomeration and a novel codebook migration strategy”, Engineering Applications of Artificial Intelligence, Volume 25, Issue 6, September 2012, Pages 1212-1225. [18] Dimitrios Tsolakis, George E. Tsekouras, Antonios D. Niros, Anastasios Rigos,” On the systematic development of fast fuzzy vector quantization for gray scale image compression”, Neural Networks, Volume 36, December 2012, Pages 83-96. [19] Hsuan-Ming Feng, Ching-Yi Chen, Fun Ye,” Evolutionary fuzzy particle swarm optimization vector quantization learning scheme in image compression”, Expert Systems with Applications, Volume 32, Issue 1, January 2007, Pages 213222. [20] Xiangwei Kong, Renying Wang, Guoping Li,” Fuzzy clustering algorithms based on resolution and their application in image compression”, Pattern Recognition, Volume 35, Issue 11, November 2002, Pages 2439-2444. [21] Abdel –Ouahab BOUDRAA,.Qosai KANAFANI,Azeddine BEGHDADI and Anissa ZERGAINOH,”Vector quantization for image compression based on fuzzy clustering”,Fifth international symposium on signal processingand its applications,ISSPA ’99,Brisbane,Australia,22-25,1999. [22] S. Makrogiannis, G. Economou, S. Fotopoulos ,”Region oriented compression of color images using fuzzy inference and fast merging”, Pattern Recognition, Volume 35, Issue 9, September 2002, Pages 1807-1820. [23] Xiaoyuan Yang, Hui Ren, Bo Li,” embedded zerotree wavelets coding based on adaptive fuzzy clustering for image compression”, Image and Vision Computing, Volume 26, Issue 6, 2 June 2008, Pages 812-819. http://www.ijettjournal.org Page 208