A Novel Compression Technique for JPEG Error Analysis and for Digital Image Applications

Transcription

1 A Novel Compression Technique for JPEG Error Analysis and for Digital Image Applications Chinta.ChandraSekhar and Ch.Ramesh 2 Student (M.Tech), Dept. of CSE,AITAM,Tekkali,A.P. csraitam@gmail.com 2 Professor, Dept. of CSE, AITAM,Tekkali,A.P. chappa_ramesh0@yahoo.co.in 84 Abstract--In computing, JPEG is a commonly used method of lossy compression for digital photography (image). The degree of compression can be adjusted, allowing a selectable tradeoff between storage size and image quality. JPEG compression is used in a number of image file formats. The compression method is usually lossy, meaning that some original image information is lost and cannot be restored.there is an optional lossless mode defined in the JPEG standard. There are so many previous methods were there for jpeg compressions which were not performing well on some errors like quantization, rounding and truncation errors. The proposed method deals about all these error minimizations by JPEG error analysis. JPEG is one of the most extensively used image formats. Understanding the inherent characteristics of JPEG may play a useful role in digital image. The main errors of JPEG include quantization, rounding, and truncation errors. Through theoretically analyzing the effects of these errors on single and double JPEG compression, we have developed three novel schemes for image including identifying whether a bitmap image has previously been JPEG compressed, estimating the quantization steps of a JPEG image, and detecting the quantization table of a JPEG image. When compared to existing techniques the proposed method outperforms for the images of small sizes. The proposed method can also reliably detect JPEG image blocks, which are as small as 8X8 pixels. The work is mainly used for digital image authentication. Extensive experimental results show that our new methods significantly outperform existing techniques especially for the images of small sizes.. Introduction With the continual expansion of multimedia and Internet applications, the needs and requirements of the technologies used, grew and evolved. To address these needs and requirements in the specific area of still image compression, many efficient techniques with considerably different features have recently been developed for both lossy and lossless compression []-[5]. The evaluation of lossless techniques is a simple task where International Journal of Latest Trends in Computing IJLTC, E ISSN: Copyright ExcelingTech, Pub, UK ( compression ratio and execution time are employed as standard criteria. Contrary, the evaluation of lossy techniques is difficult task because of inherent drawbacks associated with the objective measures of image quality, which do not correlate well with subjective quality measures. Since the mid-80s, members from both the International Telecommunication Union (ITU) and the International Organization for Standardization (ISO) have been working together to establish an international standard for the compression of grayscale and color still images. This effort has been known as JPEG, the Joint Photographic Experts Group. Officially, JPEG corresponds to the ISO/IEC international standard 0928-, digital compression and coding of continuous-tone (multilevel) still images [6]. After evaluating a number of coding schemes, the JPEG members selected a Discrete Cosine Transform (DCT) based method. JPEG became a Draft International Standard (DIS) in 99 and an International Standard (IS) in 992 []- [3]. Much research has been undertaken on still image coding since JPEG standard was established. JPEG2000 is an attempt to focus these research efforts into a new standard for coding still images. The standardization process has already produced the Final Draft International Standard (FDIS) [7]. One of the aims of the standardization committee has been the development of Part I, which could be used on a royalty and fee free basis. This is important for the standard to become widely accepted, in the same manner as the original JPEG is now. The scope of JPEG2000 includes not only new compression algorithms, but also flexible compression architectures and formats. The standard intends to compliment and not to replace the current JPEG standards. It addresses areas where current standards fail to produce the best quality or performance. JPEG2000 should provide low bit rate operation (below 0.25 bits/pixel) with subjective image quality performance superior to existing standards, without sacrificing performance at higher bit rates. Image compression scheme in JPEG2000 Part I is based on discrete wavelet transform (DWT). In this paper we attempt to evaluate and compare image quality in two

2 85 mentioned still image coding system: lossy baseline JPEG [8] and JPEG2000 image coding standard Part I [9]. JPEG and JPEG2000 use different compression techniques, which introduce different types of impairment in the reconstructed images. To describe image distortions of dissimilar nature, we used four test images with different spatial and frequency characteristics. Image quality is measured using peak signal-tonoise ratio (PSNR) [0] as most common objective measure, which does not correlate well with subjective quality measure, and picture quality scale (PQS) [], which incorporates model of human visual system (HVS) and leads to better correlation with the response of the human observers. 2. JPEG and JPEG2000 image compression techniques Because theoretical analysis of JPEG and JPEG2000 image compression techniques is widely available, in this section the main focus is given to explanation relying on experimental results. Lossy baseline JPEG is the very well-known and popular standard for compression of still images. In baseline JPEG mode, the source image is divided in 8 8 blocks and each block is transformed using DCT. The data compression is achieved via quantization followed by variable length coding (VLC). The quantization step size for each of the 64 DCT coefficients is specified in a quantization table, which remains the same for all blocks in the image. In JPEG, the degree of compression is determined by a quantization scale factor. Increasing the quantization scale factor leads to coarser quantization, this gives higher compression and lower decoded image quality. The DC coefficients of all blocks are coded separately, using a predictive scheme. The block-based segmentation of the source image is fundamental limitation of the DCT-based compression system. The degradation is known as "blocking effect" and depends on compression ratio and image content. JPEG is very efficient coding method but the performance of block-based DCT scheme degrades at high compression ratio. In recent time, much of the research activities in image coding have been focused on the Discrete Wavelet Transform (DWT) which has become a standard tool in image compression applications because of their data reduction capability [2], [3]. DWT offers adaptive spatial-frequency resolution (better spatial resolution at high frequencies and better frequency resolution at low frequencies) that is well suited to the properties of HVS. It can provide better image quality than DCT, especially at higher compression ratio [3]. JPEG2000 Part I coding procedure is based on DWT, which is applied on image tiles. The tiles are rectangular non-overlapping blocks, which are compressed independently. Using DWT tiles are decomposed into different decomposition (resolution) levels. These decomposition levels contain a number of sub bands, which consist of coefficients that describe the horizontal and vertical spatial frequency characteristics of the original tile component. In JPEG2000 Part I power of 2 decompositions are allowed (dyadic decomposition) and two types of wavelet filters are implemented: Dubieties 9-tap/7-tap filter and 5-tap/3-tap filter. Due to its better performance for visually lossless compression, the 9-tap/7-tap filter is used by default. After transformation, all transform coefficients are quantized. Scalar quantization is used in Part I of the standard. Arithmetic coding is employed in the last part of the encoding process. 3. Test Images The fundamental difficulty in testing image compression system is how to decide which test images to use for the evaluations. The image content being viewed influences the perception of quality irrespective of technical parameters of the system [4]. Normally, a series of pictures, which are average in terms of how difficult they are for system being evaluated, has been selected. We used four types of test images (52 52, 8 bits/pixel) with different spatial and frequency characteristics: Baboon, Fingerprint, Goldhill and Lena, Fig.. Characteristics of test images are evaluated in spatial domain using spatial frequency measure (SFM) [0] and in frequency domain using spectral activity measure (SAM) [2]. SFM is defined as follows: R= M N MN (x j,k x j,k ) 2, j= k= 2 Where R is row frequency, C is column frequency, x j, k denotes the samples of image, M and N are numbers of pixels in horizontal and vertical directions respectively. SAM is a measure of image predictability. It is defined as the ratio of the arithmetic and the geometric mean of the Discrete Fourier Transform (DFT) coefficients: SAM= SFM= R 2 +C 2, C= N M MN (x j,k x j, k ) 2 k= j=2 M N MN F ( j,k ) 2 j= 0 k=0 M [ N M N F ( j,k)] j=0 k=0

3 86 Where F(j,k) is (j,k)-th DFT coefficient of image. SAM has a dynamic range of [, ). Higher values of SAM imply higher predictability. Active images (SAM close to ) are in general difficult to code. These images usually contain large number of small details and low spatial redundancy. Figure.. Test images Test image Baboon has a lot of details and consequently large SFM and small SAM. Large value of SFM means that image contains components in high frequency area and small value of SAM means low predictability. It returns that Baboon presents low redundant image, which is difficult for compression. Test image Fingerprint is not typical natural image because this image has relatively large SFM but also large SAM. For typical natural image largest value of SFM implies smaller value of SAM. Image Fingerprint is easier for coder to handle than Baboon. Images Gold hill and Lena are images with less detail (smaller SFM) than Baboon. Image Gold hill has higher SFM and lower SAM than Lena. It indicates that image Lena has higher predictability than Gold hill. 4. Picture Quality Measure The image quality can be evaluated objectively and subjectively [5]. Objective methods are based on computable distortion measures. Standard objective measures of image quality are Mean Square Error (MSE) and Peak Signal-to-Noise Ratio PSNR(dB)= 0log 0( MSE) 2552 (PSNR) which is defined as for the common case of 8 bits per picture element of input image. MSE and PSNR are the most common methods for measuring the quality of compressed images, despite the fact that they are not adequate as perceptually meaningful measures of picture quality. In fact, in image compression systems, the truly definitive measure of image quality is perceptual quality [6]. The distortion is specified by mean opinion score (MOS) or by picture quality scale (PQS). MOS is result of perception based subjective evaluation described in ITU-R BT Rec. 500 [7]. PQS methodology was proposed in [] as perception based objective evaluation. It has been developed in the last few years for evaluating the quality of compressed images. In addition to the commonly used PSNR, we chose to use a perception based objective evaluation that is quantified by PQS. It combines various perceived distortions into a single quantitative measure and perfectly responds to a mean opinion score. To do so, PQS methodology uses some of the properties of HVS relevant to global image impairments, such as random errors, and emphasizes the perceptual importance of structured and localized errors. PQS is constructed by regressions with MOS, which is 5- level grading scale developed for subjective evaluation. (5-imperceptible, 4-perceptible, but not annoying, 3-slightly annoying, 2-annoying, -very annoying). 5. Method for the Image Compression Quality Comparison There are two different image compression methods: image compression with quality loss and with-out loss [7, 8]. Images compressed using first method can be re-stored to their previous state without distortion at any time. The second group of image compression methods compresses with better ratio, but there always are differences after image is restored from its compressed state. These differences can be bigger or smaller. In many image usage cases small differences are acceptable. When it is required to say whether differences are acceptable or not distortion should be measured. One of the ways to measure differences between original and compressed image is to calculate mean exponential error. This method was chosen because it accumulates exponential error from every image pixel. This means that even smallest differences will be evaluated. Mean exponential error is described by the following equation. () Where: D mean exponential error, N number of pixels in the image, A intensity of the pixel i from the original image, Agi intensity of the pixel i from the compressed image. Originally images are rectangular. For the real images, this expression must be modified. Rectangular images have two dimensions: X and Y. Modified for the two dimensional arrays () looks like this: D= X N D= N i= e A i A gi X Y Y e A ij -A gij i= 2 j= 2 (2)

4 87 where: D mean exponential error, X image width, Y image height, Aij intensity of the pixel ij from the original image, Agij intensity of the pixel ij from the compressed image. Expression (2) already can be used for the calculation of mean exponential error in gray images. But pixels of the color images require more than one value for the description of the exact color. For this purpose, in the color images, color bands are used. For example, in the recently most usable color coding system RGB there are three color bands: red, green and blue. The exact color of the image pixel is de-scribed by the combination of intensities from each color band. Gray images have only one band, RGB three. Also there are color-coding systems with four bands and other. Expression (3) adapted for color images looks like this: D= X 2 Y 2 X Y C C e A ijk-a gijk, i= j= 2 k=2 (3) where: D mean exponential error, X image width, Y image height C number of color bands, Aijk intensity of the pixel ij in the color band k from the original image, Agijk intensity of the pixel ij in the color band k from the compressed image. 6. System Analysis In this paper, we first analyze the main sources of errors introduced by JPEG compression and their relationships to JPEG recompression, and then propose three novel methods for the estimation of JPEG history. The new methods are derived based on the following observations: ) The ac coefficients of an image will increase in the range of while decrease significantly in the union regions of and after JPEG compression with quantization steps that are equal to or larger than 2. Based on this observation, a -D feature can be obtained to differentiate between uncompressed and JPEG images. 2) Due to rounding error, the ac coefficients of a JPEG decompressed image for a given frequency component will not present exactly at the multiples of the corresponding quantization step, but will spread around them at an interval between with a high probability. By removing such rounding effect, an effective feature can be obtained for quantization step estimation. 3) When a JPEG image is JPEG compressed again, most original pixels will be better preserved if the recompression uses the same quantization table as that used for the original JPEG. This is true even when the recompression quality factor is set to 00 (all quantization steps are equal to ). Based on this observation, we obtain a simple but very effective method to extract the quantization table from a JPEG decompressed image. We present theoretic analysis and experimental results which show that our techniques are very effective and perform better than existing methods, especially for the images of small sizes. 7. Error Analysis for Jpeg Compression and Recompression Lossy JPEG compression is a block-based compression scheme. Each non-overlapping 8X 8 block within an image is Figure. 2. Single and double JPEG compression and decompression. (a) Single JPEG; (b) double JPEG dealt with separately. In the following, we will consider the processes of JPEG compression and recompression on an 8X 8 uncompressed block. As shown in Fig. 2(a), the image block B is first transformed from the spatial domain into the frequency domain using DCT, and then the frequency coefficient d is quantized by a quantization table. The error introduced in this stage is called quantization error, the main cause of information loss for the images having not been JPEG compressed before. The quantization coefficient Qd is then further compressed by the lossless entropy encoding. Finally, the resulting bit stream is combined with a header to form a JPEG

5 88 file. In JPEG decompression, the compressed file is first entropy decoded to recover the quantized coefficient Qd exactly, which are then multiplied by the table Q to obtain the DE quantized coefficient. And then the inverse DCT (IDCT) is performed on.to get the pixel values in realvalued representation. Since gray-scale images are typically stored with 8 bits per pixel, which can represent 256 different intensities, the resulting real values will be rounded and truncated to the nearest integers in the range of [0, 255]. Two errors will be generated in this stage, namely rounding and truncation errors. Please note that the rounding error usually occurs in every 8X8 block, while the truncation error does not. 8. Results Case study : Case study 2: Enter image processing size: 28 Enter the quality factor (QF):= 95 Quantization Matrix used Processing JPEG- coding... Processing JPEG-2 coding... Err = Case study 3: Enter image processing size: 64 Enter the quality factor (QF):= 98 Quantization Matrix used Processing JPEG- coding... Processing JPEG-2 coding... Err = Enter image processing size: 256 Enter the quality factor (QF):= 78 Quantization Matrix used

6 89 Processing JPEG- coding... Processing JPEG-2 coding... Err = Conclusion In this paper, we have focused on the problem of estimating primary quantization steps for selected low-frequency DCT coefficients. However, in practice we may need the entire quantization matrix. Because coefficients of higher frequencies are more often quantized to zero than low-frequency coefficients, we believe that reliable estimation of higher frequency coefficients is not possible due to insufficient statistics. Results of our experiments support this expectation. We recommend first estimating selected low-frequency quantization steps and then estimate the rest of the primary quantization matrix from some standard matrix whose low-frequency steps are close (identical) to the estimated coefficients. References [] ACR/NEMA. DICOM: Digital Imaging and Communications in Medicine. Rosslyn, NEMA publications, 200. [2] J. Riesmeier, M. Eichelberg, D. Lemoine, V. Punys, N. Balogh, P. Jensch, J. Punys. DICOM Extensions for Narrow-Band Networks. Information technology and control, Kaunas, Technologija, 200, No.4 (2), [3] Image file formats. Wikipedia, the free encyclopedia. Available at: formats. Accessed June 6, [4] D. Hankerson, G.A. Harris, P.D. Johnson, Jr. Intro-duction to Information Theory and Data Compression (Second Edition). CHAPMAN&Hall/CRC Press Com-pany, [5] P. Fränti, O. Nevalainen, T. Kaukoranta. Compres-sion of Digital Images by Block Truncation Coding: A survey. The Computer Journal, Vol.37, No.4, 994, [6] G.K. Wallace. The JPEG Still Picture Compression Standard. Communications of the ACM, Vol.34, No.4, 99, [7] P.Cosman, R. Gray, R. Olshen. Evaluating quality of compressed medical images: SNR, subjective rat-ing, and diagnostic accuracy. Proceedings of the IEEE 82, 994, [8] A.M. Eskicioglu, P.S. Fisher. Image quality measures and their performance. IEEE Transactions on Commu-nications 43, 995, [9] JPEG2000, World Wide Web: [0] M. Eskicioglu, P. S. Fisher, Image Quality Measures and Their Performance, IEEE Transactions on Communications, Vol. 43, No. 2, December 995, pp [] M. Miyahara, K. Kotani, V. R. Algazi, Objective Picture Quality Scale (PQS) for Image Coding, age?96-2 [2] M. Antonini, M. Barland, P. Mathieu, I. Daubechies, Image Coding Using the Wavelet Transform, IEEE Trans. on Image Processing, Vol., 992, pp [3] S. Grgic, K. Kers, M. Grgic, Image Compression Using Wavelets, Proceedings of the IEEE International Symposium on Industrial Electronics, ISIE'99, Bled, Slovenia, 999, pp [4] S. Bauer, B. Zovko-Cihlar, M. Grgic, The Influence of Impairments from Digital Compression of Video Signal on Perceived Picture Quality, Proceedings of the 3rd International Workshop on Image and Signal Processing, IWISP'96, Manchester, 996, pp [5] J. Allnatt, Transmitted-picture Assessment, John Wiley and Sons, 983 Author Chinta Chandrasekhar received B. Tech degree fromjntu, Hyderabad, India. He is currently pursing (M. Tech) degree in Computer Science & Engineering from JNTU, Kakinada.His area of interests include Image Processing, Data Mining and Network Security. Ch. Ramesh received the B. Tech degree from Nagarjuna University, Nagarjuna Nagar, India and the M. Tech degree in Remote Sensing from Andhra University, Vishakapatnam, and another M. Tech degree in Computer Science & Engineering from JNTU, Hyderabad. He is currently pursuing the Ph. D degree in Computer Science from the University of JNTU, Hyderabad.He is a professor in the department of Computer Science & Engineering, Aditya Institute of Technology and Management, Tekkali, India. His research interests include Image Processing, Pattern Recognition, and Formal Languages.