http:^^www.cs.cornell.edu^info^people^mukai^cs631^jpeg.html
来自「This data set contains WWW-pages collect」· HTML 代码 · 共 301 行
HTML
301 行
MIME-Version: 1.0
Server: CERN/3.0
Date: Sunday, 01-Dec-96 19:27:41 GMT
Content-Type: text/html
Content-Length: 14748
Last-Modified: Thursday, 18-Jan-96 18:26:34 GMT
<HTML><head><title> JPEG Encoding Using Perceptual Quality </title> </head><body BGCOLOR="#FFFFFF"><IMG SRC="Include/lena.gif" border=0 align=left><h2><CENTER>Multimedia Systems<br>JPEG Encoding using Perceptual Quality<br></CENTER></h2> <center> Nobuhiko Mukai (mukai@cs.cornell.edu)<br> Lucy Y. Wu(yuwu@cs.cornell.edu)<br> Mikio Sakurai (msakurai@sunlab.cit.cornell.edu)<br></center> <img src="Include/line_rain.gif"><h4>Abstract:<br>The key point of image compression is to achieve a low bit rate in the digital representation of an input image or video signal with minimum perceived loss of picture quality. Over the past several years there have been many attempts to incorporate perceptual masking models into image compression system. Based on the "pre-quantization", we developed two new models. Compared to JPEG default setting, both of our models significantly lower the bit-rate and keep the almost sameimage quality.<p></h4><img src="Include/line_rain.gif"><h2>1. Introduction</h2>The key point of image compression is to achieve a low bit rate in the digital representation of an input image or video signal with minimum perceived loss of picture quality. Since the ultimate criterion of quality is judged or measured by the human receiver, it is important that the compression(or coding) algorithm minimizes perceptually meaningful measures of signal distortion.<p>JPEG default quantization tables(QT) are based on psychovisual thresholding andare derived empirically. But because only one QT is available for every image,the default QT is image independent and can not be used to achieve the optimizedcompression result for each specific image.<p>Some perceptual models have been developed to calculate the image dependent QT. Butbecause every block in the image contributes different properties to the total image,one QT that is best for the whole image is not always best for every block. Ifwe can quantize every block using different QT specifically suitable for thatblock, we can get the most optimized compression result for every block.<p>Because JPEG allows only one QT for each image, pre-quantization is proposed byJohnston and Safranek(J&S)[1]. For every block, one specific masking threshold iscalculated and used to zero out the perceptual irrelevant coefficients while theother coefficients passed unchanged. Then one base QT will be finally used toquantize for all the remaining coefficients in each block.<p>Despite the benefits of simple implementation, the J&S model has the disadvantage ofcomputing a single masking elevation for each input block. This means that thereis no information about the distribution of energy within the block.<p>Thisproblem can be overcome by applying the contrast masking model for each DCTcoefficient(model-1).<p>The computation for model-1 is somewhat complex. This led us to design a secondone(model-2) based on the luminance ratio of block to total image to replace the J&Smodel. This second model has the same single masking elevation problem, but the calculationis more simple.<p>Our test result shows, compared to the JPEG default setting, both of our modelsreduced the bit rate 10% with little or no perceived loss in quality.<p>The rest of this paper is organized as follows. Section 2 describes algorithms to"prequantize" a JPEG image. Section 3 describes two perceptual models designed by us. Section 4 describes the detailed evaluation of our models, andsection 5 reviews related work and future extensions.<p><h2>2. Prequantization</h2>2.1 Quantization<br> The Baseline JPEG encoder consists of three major components, a Forward DCT,Quantization, and Entropy Coding. The step of the quantization is to take the rawoutput of the DCT and quantize the coefficients by dividing it, coefficient bycoefficient, by QT, and rounding to the nearest integer. The purpose of quantizationis to achieve compression by representing DCT coefficients with no greater precisionthan is necessary to achieve the desired image quality. In other words , the goal ofthis processing step is to discard information which is not visually significant.<p>2.2 Perceptual Model<br>Many studies have attempted to derive a computational model of this visual maskinglevel. For each block in the input image, the model attempts to determine to whatdegree the features present in that block inhibit the visual system from thedistortion introduced by the compression/decompression process. From these points, it is possible to determine a masking threshold for each DCT coefficient.<p>2.3 J&S's model<br> Johnston and Safranek have developed a framework for computing a locally adaptivemasking model based on an engineering framework. They assume that the total maskinglevel for any block of the input can be represented as a base masking level, andother multiplicative elevation factors that represent the contribution of inputdependent properties of the visual system to the total mask. This model may beexpressed as:<p>M(u,v) = Global(u, v) x Local(u, v) --- (1)<p>where M(u, v) is the masking level for frequency (u, v) of the input block,Global(u,v) is the base masking level which depends only on global properties, andLocal(u, v) handles the image specific local variation in the masking threshold. Theadaptation is derived as a function of the block standard deviation using thefollowing formula:<p><img src="Include/Fig1.xbm"> ---(2)<p><h5>Figure 1. J&S's model</h5><p>This applies to all of the AC coefficients. The masking elevation for the DCcoefficient is always set to unity. This model has the advantage of simpleimplementation, and works well in practice. It has the disadvantage of computing asingle masking elevation for each input block.<p>Figure 2 illustrates the structure of such an encoder. The forward transformation is identical to the one in baseline JPEG. At this point, the DCT coefficients are input to the perceptual model which generates the data dependent quantization table for that block. This table and the raw DCT coefficients are now input to a"pre-quantizer". The purpose of this module is to zero out the coefficients thathave a magnitude less than the corresponding entry in the quantization table forthat block, and pass the other coefficients through unchanged. Finally theseprequantized coefficients are quantized and entropy coded as in standardJPEG.<br><p> In the dequantization step, only the base QT is used.<p><img src="Include/Fig2.xbm"><p><h5>Figure 2. a structure of an encoder with prequantization</h5><p><h2>3. Our Models</h2>Our models intends to take advantage of the above prequantization method.<p><h4>3.1 Model-1</h4> The J&S's model uses the perceptual model but does notconsider the distribution of the energy within the block when calculating theblock-specific masking threshold. To overcome this problem, we employ a visualmasking that has been widely used in vision models, based on work by Legge andFoley[2]. Given a DCT coefficient c(u,v) and a luminance threshold t(u,v), themasked threshold M(u,v) is <br><p>M(u,v)= MAX[t(u,v), |c(u,v)|^w(u,v) *t(u,v)^(1-w(u,v))] ---(3)<p>w is a constant that lies between 0 and 1. When w=0, no masking occurs, and when w=1, we have "Weber's Law" behavior. For our experiment, an empirical value of w =0.7 was used.<p> In our implementation to calculate the masking threshold, we did not calculate the luminance threshold by using Peterson's model[3] as suggested by Watson[4]. In addition, as indicated in JPEG standard, JPEG default quantization tables are based on psychovisual thresholding and are derived empirically. If it is divided by 2, the almost indistinguishable image can be reconstructed. That means the default QT can be treated as a general luminance threshold. So, we replaced the t(u, v) in equation(3) by "JPEG default QT value" / 2.<p>As a Global masking level, we used default JPEG QT.<p><h4>3.2 Model-2</h4> The masking threshold computation for model-1 is rather complex. This led us to design the second model(model-2) based on the luminance ratio of block to total image. This model has the same single masking elevation problem, but the calculation is more simple.<p>Basically, model-2 follows the J&S masking model (equation 1). But for the calculation of multiplicative elevation factors (Local(u,v)), we use the luminance ratio of each block to the total image.<p>Well known "Weber's Law" is expressed as:<p>df / f = constant (0.02) ---(4)<p>f is luminance and df is the just noticeable difference.<p>This equation means our perception is sensitive to luminance contrast rather than the absolute luminance values themselves. At a given luminance f, if the block's luminance is only a little bit differ from luminance f, then the block is less visible and we can drop a lot of perceptually unimportant information.<p>The adaptation is derived as a function of the ratio of the block's average luminance to the luminance average of the total image using the following formula:<p><img src="Include/Fig3.xbm"> ---(5)<p><h5>Figure 3. model 2: masking model based on luminance ratio</h5><p>Maximum threshold elevation "max_e" and Minimum Threshold (Minimum luminance ratio) "min_t" are the parameters we need to experiment for.<br> As a Global masking level, we used default JPEG QT.<p><h2>4. Image Quality and Bitrate</h2>We compared the bitrate(bits/pixel) and Image quality of our model-1 and model-2 with those of the Baseline JPEG. For model-2, first we need to optimize the parameter. We worked on several images and found thatthese two parameters(max_e, min_t) are not strongly sensitive to SNR.To the different value of luminance ratio parameter min_t, SNR is almost constant.Threshold elevation parameter max_e has weak relation with SNR and if it gets bigger, SNR becomes worse. This feature is common to all images. Table 1 shows a result we got on image "Lena".<p><img src="Include/Table1.xbm"><p><h4>Table 1: Parameter(max_e,min_t) dependence of SNR</h4><p> For the following experiment, we choose the preferable value max_e = 2 and min_t = 0.01 for each parameter of model-2. Five images were used in this experiment: a photo of human face("Lena"), a flower scene("flowers"), a photo of animal face("baboon"), two photo of airplane(F-18 and Pitts).<p>Bitrate is calculated after compressing the image file using the pack(huffman coding) program. For the evaluation of the image quality, we used two evaluation method. The first is SNR(signal to noise ratio). In order to provide further insight into the subjective quality of the models, we used DCON metric[5]. This algorithm takes as input two images, a reference and a test, compare the difference of luminance pixel by pixel. The formula is as follows:<p>DCON = 1/N * sum((y1-y2)/(y1+y2) + 23) --- (6)<p>This method is simple but very competitive with other complicated human eyemodel metric.<p> Table 2 summarize the results of these experiments.<p><img src="Include/Table2.xbm"><p><h4>Table 2: Image quality(SNR,DCON) vs Bitrate</h4><p> Both our model-1 and model-2 compress 10% better than the Baseline JPEGwith almost no perceptual loss in quality. Figure 4 shows one example ("flowers") of output image of our models in comparison with original image and that of Baseline JPEG.<p>(1)<img src="Include/flowers_org.gif">(2)<img src="Include/flowers_jpg.gif"><p>(3)<img src="Include/flowers_model1.gif">(4)<img src="Include/flowers_model2.gif"><p>(1) original image (2) Baseline JPEG (3) model 1 (4) model 2 <br><h5>Figure 4. a sample image comparison</h5><p><h2>5. Related work and Conclusions</h2> There have been many attempts to incorporate perceptual masking model into image compression systems[6,7]. Johnston-Safranek model[1] and Watson[4] model are perhaps the most well known perceptual model. J&S model has investigated locally optimized prequantization table. Watson model has investigated image dependent masking model which incorporates not only global conditions but also accounts for local contrast masking.<p>Klein of U.C.Berkery optometry school has reported techniques for improving the quality of JPEG with high compression rate in viewpoint of vision community[8]. He suggests that with improved human vision models the quantization step could be made more effective by considering effects such as mean luminance, color, bandpass filters in spatio-temporal frequency and orientation, contrast masking and human contrast sensitivity.<p>The primary contribution of our work is that it details encoding-specificprequantization algorithms that compress images at high compression rate with minimal artifact. Our future work will include extending this work to include other human eye related factors such as spatio-temporal frequency and orientation.<p><img src="Include/line_rain.gif"><h3> References </h3>[1]Johnston, J. D. and Safranek, R. J., "A Perceptually Tuned Sub-band Image Coder With Image Dependent Quantization and Post-quantization Data Compression." Proc. ICASSP 89, Glasgow, Scotland, vol.MA, May 1989, pp. 1945-1948<p>[2]G.E.Legge and J.M.Foley, "Contrast masking in human vision", Journal of the Optical Society of America. 70(12), 1458-1471(1980).<p>[3]A.J.Ahumada and H.A.Peterson, "Luminance-Model-Based DCT quantization for color image compression", SPIE:Human Vision, Visual Processing, and Digital Display III, Vol.1666,1992, PP365 - 374.<p>[4]Watson,A.B., "DCT quantization matrices visually optimized for individual images", SPIE:Human Vision, Visual Processing, and Digital Display IV, Vol 1913 Feb.1993, pp202 - 216.<p>[5]Daniel R. Fuhrmann, John A. Baro, and Jerome R. CoxExperimental evaluation of psychophysical distortion metrics for JPEG-encoded images, SPIE:Human Vision, Visual Processing, and Digital Display, Vol.1913, PP179 - 190.<p>[6]S.J.Daly, "The visible difference predictor: an algorithm for the assessment of image fidelity",SPIE:Human Vision, Visual Processing, and Digital Display III, Vol.1666,1992, PP2 - 15.<p>[7]J.L.Mannos and D.J.Sakrison, "The effects of a visual fidelity criterion on the encoding of images", IEEE Transaction on Information Theory IT-20, pp525 - 536<p>[8]Stanley A. Klein, Amnon D. Silverstein and Thom Carney, "Relevance of human vision to JPEG-DCT compression", SPIE:Human Vision, Visual Processing, and Digital Display III, Vol.1666,1992, PP200 - 215.<p><br> <img src="Include/blueball.gif"><A HREF="paper.ps">postscript file<img src="Include/ps.gif"></A><h5>File last modified - <EM> 4 December 1995</EM></h5> </body> </HTML>⌨️ 快捷键说明
复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?