Color feature based image compression.
Image compression is the technique used to reduce the quantity of data used to represent a file, image or video content without excessively reducing the quality of the original data. The purpose of image compression is to reduce the redundancy present in the image, so that it can be stored and transferred efficiently. The compressed image has less number of bits compared to original. Hence, the required storage space will be reduced, consequently more images can be stored and it can transfer fast to save the time, transmission bandwidth. In this work, we have identified such problems and tried to provide an effective solution to these problems.
1. Related Works:
Neelamma K. Patil et al proposed an adaptive method for image compression which results in better compression and reduce storage space.
Gaurav Vijayvargiya et al analyzed the various techniques of Image Compression. Finally they found that lossless compression techniques are more effective over lossy compression.
Nitesh Kumar More et al presented a JPEG Picture Compression Using Discrete Cosine Transform. This system compresses an image by DCT which separates the image into different frequency and quantize the frequencies where part of compression occurs and then entropy encoding. Finally a reconstructed image is obtained with compression ratio ranging from 88% to 96%.
Yung-Gi Wu proposed an adaptive sampling algorithm by calculating the difference area between correct points and predicted points to decide the significant coefficients from DCT coefficients. Experimental results show that higher compression ratio compared to JPEG compression.
Dong ping Tian presented a review on Image Feature Extraction and Representation Techniques and analyses the effectiveness of the fusion of global and local features in image processing.
Nageswara Rao Thota et al, presented an Image Compression Using Discrete Cosine Transform. The system consists of three stages l.transform; 2.Quantization; 3.Coding.Restored image with different coefficient are observed and finally produce an 8X8 compressed image.
Mamta Sharma analyzed Huffman algorithm and compares it with other common compression techniques. Finally concludes that Huffman algorithm is used in JPEG compression which produces optimal and compact code.
Bartkowiak, M., Doman ski, M., described a very simple technique for compression of chrominance data in color images and video sequences coded at low bit rates. An experimental result proves that this technique is very efficient for high compression of chrominance.
The proposed block diagram is shown in Fig.1.In this paper, the color and texture feature are extracted to select significant DCT coefficient which reduces the bandwidth and storage space. The RGB image is converted into YCbCr. The planes are separated and individually the plane divided into 8X8 blocks and DCT is applied to each block. Then, Quantization process using Q50 quantizer. Features are extracted from each block of Y plane, Cb plane and Cr plane which select significant DCT coefficients. The Zigzag scan is used to convert matrix into an array. Huffman encoding is used to convert coefficients into binary values and transmitted. The transmitted binary values are decoded. Dequantized and decompressed using IDCT. Finally, YCbCr to RGB is applied to obtain reconstructed image.
3. Color Conversion:
The original RGB image to YCbCr color space conversion is performed to extract texture and color features from Y plane, Cb plane and Cr plane. The formulae for converting from RGB to YCbCr are given below.
Y = (77/256) R + (150/256) G + (29/56)B (1)
Cb = -(44/256)R - (87/256)G + (131/2B6)B + 12B (2)
Cr = (13/2B6)R - (110/256)G - (21/2B6)B + 12B (3)
The image compression works best on the luminance and chrominance color space. The human eye has different sensitivity to color and brightness. Thus there came about the transformation of RGB to YCbCr. Y is called the luminance and Cb, Cr are called the chrominance.
4. DCT (Discrete Cosine Transform):
The three color components of the YCbCr image are divided into many 8*8 blocks and DCT is applied to each block. A DCT expresses a sequence of finitely many data points in terms of a sum of cosine functions oscillating at different frequencies. The definition of 2-D DCT is given by,
C(u, v) = [alpha](u)[alpha](v)[[SIGMA].sup.N-1.sub.x=0] [[SIGMA].sup.N-1.sub.y=0] f(x,y) cos ([pi](2x + 1)u/2N) cos ([pi](2x + 1)v/2N) (4)
Where u,v=0,l,2,......., N-1 and
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Next, DCT coefficients are compressed through quantization process using Q50 quantizer. Quantization is the step where we actually throw away data.
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)
During quantization every coefficients in the 8*8 DCT matrix is divided by a corresponding quantization matrix values and then rounding to the nearest integer value. The goal of quantization is to reduce most of the less important high frequency DCT coefficients to zero, the more zeros we generate better the image will compress. After the process of quantization, most of DCT coefficients become zero.
6. Feature Extraction:
Feature extraction plays a prominent role in processing an image by describing image with set of features rather than pixels.
Color Feature- The color features are extracted from chrominance components namely Cb and Cr planes from YCbCr color space using mean and standard deviation.
Mean represents some brightness of the image. mean = [mu] = 1/N[[SIGMA].sup.N-1.sub.i=0][x.sub.i] (6)
Standard deviation reveals something about the contrast of image.
standard deviation = [sigma] = [square root of 1/N [[SIGMA].sup.N.sub.i=0][([x.sub.i] - [mu]).sup.2]] (7)
Where Xi - number of elements in the sample and
N - Sample size.
Texture Feature-Texture is one of the important characteristics of an image. Texture features are extracted from Luminance component namely Y plane from YCbCr color space. 14 texture features are extracted from the Co-occurrence matrix. Four GLCM texture features are commonly used are:
Contrast measures of the intensity contrast between a pixel and its neighbour over the whole image.
Contrast = [E.sup.m-1.sub.k=0][k.sup.2][[SIGMA].sub.[absolute value of i - j]=k]M(i,j) (8)
Energy measures of the extent pixels pair repetition (uniformity).
Energy = [[SIGMA].sub.i] [[SIGMA].sub.j][(M(i, j)).sup.2] (9)
Homogeneity measures homogeneity as it assumes larger value for smaller gray tone difference in pair.it is also known as inverse difference moment.
Homogeneity = [[SIGMA].sub.I,j.sup.M(I,j)]/1 [+ or -][absolute value of i - j]) (10)
Entropy measures of the uniform distribution of levels.
Entropy = [[SIGMA].sub.i] P * log P (11)
Where M(i,j) - input image,
i - Row number and
j - Column number.
7. ASDC (Adaptive Significant DCT Coefficient):
After Color and Texture features for each block are extracted, these features are used for the selection of significant DCT coefficients. Sufficient significant coefficients are selected to retain color and texture information.
8. Zig-Zag Scan:
After the quantization, the nature of the quantized DCT coefficients and the random occurrence of zeros in the high frequency coefficients lead to the further need of the compression. The purpose of the Zig-zag Scan is to group low frequency coefficients are in top of vector.
9. Huffman Encoding:
After selection of significant DCT, Huffman encoding is performed to convert coefficients into binary digits.
Steps for Huffman algorithm,
1. Find the probabilities for each source symbols to be coded.
2. Order the probabilities of the symbols in descending order.
3. Next, adding the lowest two probabilities into single and assign zero & one to left and right branches, respectively.
4. If there are more than one unmerges node, repeat step 2 and 3 otherwise read bits on the branches from top to bottom to generate code words.
10. Reconstruction of Image:
The transmitted binary digits are decoded at the receiving end. Decoded information is then dequantized. Then, decompression using Inverse DCT to obtains the reconstructed image. The definition of 2-D IDCT is given by,
f(u, v) = [[SIGMA].sup.N-1.sub.x=0][[SIGMA].sup.N-1.sub.x=0][alpha](u)[alpha](v)C(x, y) cos([pi](2x + 1)u/2N) cos ([pi](2x + 1)v/2N) (12)
Where u, v=0,l,2,.......,N-l and
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Color conversion from YCbCr to RGB is performed to convert reconstructed image into original RGB color image.
Experimentation is performed on image formats namely .bmp, .png and .tif. Few sample input images are shown in fig.2.
12. Experimental Results:
Fig.3.show the reconstructed image of Fig.2. In reconstructed image there is no noticeable loss of information for human eye to recognize in reconstructed image.
13. Performance Measure:
The quality of image is measured in terms of Mean Squared Error (MSE) and Peak Signal to Noise Ratio (PSNR) using equations,
MSE = 1/mn [[SIGMA].sub.i][[SIGMA].sub.j][[M(i, j) - (R(i,j)].sup.2] (13)
Where M(i,j) - input image,
R(i,j) - reconstructed image and
m,n - no.of rows and columns.
PSNR = 20 * log(255/[square root of MSE])(dB) (14)
MSE and PSNR for proposed method are tabulated in Table.1. The number of coefficients used in the proposed method is much lesser than total no. of coefficients.
Image formats like. tif, bmp and .png are used for testing and their compression ratio is shown Fig. 4. Compression ratio achieved is above 80% for different image formats.
The objective of this paper is to achieve good compression ratio by retaining color and texture features of an image and to utilize storage space and bandwidth efficiently during transmission. The problem of encoding all DCT coefficients which requires huge band width and storage space is reduced by selecting significant DCT coefficients using feature extraction process. Experimental results show that the proposed image compression system gives good compression ratio and good quality of reconstructed image.
Received 12 October 2014
Received in revised form 26 December 2014
Accepted 1 January 2015
Available online 17 February 2015
Bartkowiak, M., M. Domanski, 2000. High compression of chrominance data by use of segmentation of luminance. Proceedings of European Signal Processing Conference EUSIPCO'OO, 2.
Dong ping Tian, 2013. A Review on Image Feature Extraction and Representation Techniques. International Journal of Multimedia and Ubiquitous Engineering, 8: 4.
Gaurav Vijayvargiya, Dr. Sanjay Silakari, Dr.Rajeev Pandey, 2013. A Survey: Various Techniques of Image Compression. (IJCSIS) International Journal of Computer Science and Information Security, 11: 10.
Mamta Sharma, 2010. Compression Using Huffman Coding. IJCSNS International Journal of Computer Science and Network Security, 10: 5.
Nageswara Rao Thota, Srinivasa Kumar Devi reddy, 2008. Image Compression Using Discrete Cosine Transfonn.Georgian Electronic Scientific Journal: Computer Science and Telecommunications, 3.
Neelamma, K., Patil, Suresh F. Murgod, Lokesh Boregowda, V.R. Udupi, 2013. Adaptive Texture and Color Feature Based Color Image Compression. 2013 International Conference on Smart Structures & Systems (ICSSS-20 13).
Nitesh Kumar More and Sipi Dubey, JPEG Picture Compression Using Discrete Cosine Transform. International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064.
Yung-Gi Wu, 2002. Medical Image Compression by Sampling DCT Coefficients. IEEE Transactions on Information Technology in Biomedicine, 6: 1.
(1) Dr. R. PRIYA and (2) B. ANUBALA
(1) Associate Professor, Annamalai University, Chidambaram, Tamil Nadu, INDIA.
(2) PG Student, Annamalai University, Chidambaram, Tamil Nadu, INDIA.
Corresponding Author: Dr. R. PRIYA, Associate Professor, Annamalai University, Chidambaram, Tamil Nadu, INDIA.
Table 1: Computation of MSE, PSNR, no. of coefficients and total no. of coefficients Image MSE PSNR No. of Coeff. Total No. during of Coeff. transmission I1.tif 2.3218 44.5401 18872 230266 I2.tif 17.2459 35.7708 39232 232474 I3.tif 6.3321 40.1696 22182 232268 I4.png 13.1794 36.9426 35067 156497 I5.png 14.0833 36.7426 32155 170025 I6.png 6.0416 40.3588 24847 128322 I7.bmp 6.1174 40.5273 27324 132508 I8.bmp 9.2656 38.5321 25362 132562 I9.bmp 4.8799 41.4526 24386 125081 Fig. 4: Bar Graph of Compression ratio (CR in %) Image compression I1.tif 92.3141 I2.tif 98.916 I3.tif 97.6854 I4.png 87.8701 I5.png 85.4069 I6png 80.8848 I7.bmp 87.2332 I8.bmp 80.5872 I9.bmp 84.5604 Note: Table made from bar graph.
|Printer friendly Cite/link Email Feedback|
|Author:||Priya, R.; Anubala, B.|
|Publication:||Advances in Natural and Applied Sciences|
|Date:||Jun 1, 2015|
|Previous Article:||An intelligent face and finger recognition using median filtering.|
|Next Article:||Detection and recognition of real-time indoor objects using sift features.|