《基于相关因素映射的小波回归分析短期负荷预测模型研究.》由会员分享，可在线阅读，更多相关《基于相关因素映射的小波回归分析短期负荷预测模型研究.（9页珍藏版）》请在金锄头文库上搜索。
1、 Improved DCT-domain Color Image Enhancement through Anisotropic Diffusion Yi Wan, Yafeng Ma Institute for Signals and Information Processing Lanzhou University Lanzhou, P. R. China 730000 Abstract Color image enhancement directly in the DCT domain is an efficient approach in high speed data transmi
2、ssion. However, such methods often suffer from blocking artifact due to inappropriate scaling between adjacent blocks. We propose an improved method based on anisotropic diffusion that overcomes this difficulty. Simulation results show that the proposed method can effectively remove blocking artifac
3、t introduced during the DCT coefficient scaling stage, resulting in much improved performance over some of state of the art methods. Keywords: DCT, color image enhancement, blocking artifact, anisotropic diffusion. 1 Introduction Image enhancement is a class of effective post-processing techniques t
4、o improve the visual quality of commonly obtained images 1, 2. The general approach is to dynamically adjust the (local) range of image values so as to increase the contrast and other aspects of visual perception quality. Most image enhancement methods are performed in the spatial domain. With incre
5、asing number of images in compressed format, compressed domain image enhancement algorithms are also proposed recently (e.g., 3, 4, 5, 6, 7), thus eliminating the need for decompression and compression. The majority of the current image enhancement research is focused on grayscale images. For color
6、images, the traditional approach are to fi rst obtain the intensity component by transforming the pixel values from space to, for example, the space or space. Then perform enhancement on the intensity image as in the case of grayscale images. Recently it is pointed out in 1 that this approach could
7、cause color distortion because the colors may not be preserved in the enhanced image, and a very efficient method called the CES (Color image Enhancement by Scaling) algorithm is proposed in which all the DCT coefficients within a processing block are scaled by a constant factor. This has produced v
8、ery good enhancement performance. Yet we fi nd that the commonly encountered blocking artifact still remains, even with the suppression technique in 1. We propose a more effective algorithm for blocking artifact removal based on anisotropic diffusion. Simulation results show that it achieves favorab
9、le performance over other algorithms such as those in 1 and 3. RGB HSV rbC YC The rest of the paper is organized as follows. In Section 2 we review the algorithm in 1 and analyze its effect on blocking artifact. Another recent blocking artifact removal method in 3 is also briefl y reviewed. -1- 中国中国
10、科技论文在线科技论文在线 In Section 3 we present the proposed improvement, which keeps the enhancement framework of the CES algorithm in 1 but replaces its blocking artifact suppression part by an anisotropic diffusion process. In Section 4 we present the enhancement results of the proposed algorithm and compar
11、e its blocking artifact removal performance with other methods. Finally, we make conclusions in Section 5. 2 Review of the color image enhancement by scaling the DCT coefficients We fi rst review the CES algorithm in 1. Then we analyze its performance on blocking artifact suppression. 2.1 The CES al
12、gorithm In the CES algorithm, for each 88 block (the block size can be NN in general), the DCT coefficients , , of the Y, , components are fi rst obtained and the block is processed independently of the others. A scaling factor c Y c U c V b C r C is initially determined according to a mapping curve
13、 such as the S-curve (The block pixel maximal value is used for normalization.) 1, 8. Then it is clamped above and below according to max I /( max ),min(B and ) 1 ,max(, where is the highest possible pixel value, max B and are the mean and standard deviation of the block pixel values (Note that and
14、can be computed directly in the DCT domain 1), and is a constant parameter. In 1 it is proved that if the block pixel values lie within the range of , and )/(, max 1B c U c V , then the enhanced pixel values will not exceed . Finally the DCT coefficients , , are scaled by max B c Y . 2.2 Analysis on
15、 the blocking artifact The basic algorithm in the above subsection will encounter the common problem of blocking artifact. Therefore 1 introduces the following approach: If the current block tends to be blocking, then subdivide it into four sub-blocks of size 44 and process these four blocks in the
16、same way. In this approach, the key is to determine when to subdivide a block. In 1, a threshold thresh is fi rst set up. If thresh , then perform block subdivision. Since the magnitude of indicates the high frequency content in the block, the idea behind this approach is that we keep the block if it is generally smooth and subdivide it if it is not. Next we show that such a rational is not generally valid. In Figure 1(a) we
