📄 http:^^www-cse.ucsd.edu^users^jhan^hc.html
字号:
Date: Wed, 15 Jan 1997 01:31:37 GMTServer: NCSA/1.4.2Content-type: text/htmlLast-modified: Tue, 10 Dec 1996 04:47:30 GMTContent-length: 8553<HTML><HEAD><TITLE>Hierarchical JPEG</TITLE><BODY><h2>Hierarchical Coding </h2>To investigate the effectiveness ofhierarchical coding (or layered coding) for congestion controlin ATM Networks, I have implemented the sequential DCT-based hierarchical mode of JPEG using the existing baseline JPEG code from the IndependentJPEG group. Below is a short tutorial on the hierarchical mode of JPEG.<P><MENU><LI><!WA0><A href="#overview">The JPEG Still Image Compression Standard Overview</A> <LI><!WA1><A href="#hmode">The Hierarchical Mode of the JPEG Standard</A> <LI><!WA2><A href="#diff">Difference between Progressive Mode and Hierarchical Mode</A> <LI><!WA3><A href="#refer">References</A> </MENU><P><HR><h3><A NAME="overview">The JPEG Still Image Compression Standard Overview</A></h3>The JPEG (Joint Photographic Experts Group) standard was developed under the auspices ofISO (ISO 10918-1 JPEG Draft International Standard) and CCITT (CCITT Recommendation T.81), and supports both lossy and lossless compression. The lossy methods are based on the Discrete Cosine Transform (DCT). The standard specifies four modes of operation: sequential, lossless, progressive, and hierarchical encoding<!WA4><A href="#ref1">[1]</A>.The progressive and hierarchical modes allow for decompression of a partially received signal.<br><br>Even though this standard was developed with still images in mind, it is also used for video transmission by providing intraframe compression only (often referred to as motion JPEG). Intraframe compression provides much lower compression ratios for video than combined intra and interframe compressio frames. <br><br>Sequential encoding is probably the most common mode for most applications. It involves encoding each image component in a single left-to-right top-to-bottom scan. The most widely used mode of JPEG is the baseline mode which is a subset of the sequential encoding based on the Discrete Cosine Transform (DCT). In the baseline encoding algorithm, each component of the source image is divided into 8x8 pixels of non-overlapping blocks. The pixel values in each such block are first level-shifted from unsigned to signed integers and then input to the forward DCT. The resulting 64 DCT coefficient values can be regarded as the relative amount of 2D spatial frequencies contained in the input image. The DC coefficient is a measure of the average value of the 64 image pixels.<br><br>The next step is to quantize the DCT coefficients. The purpose of this step is to achieve further compression by quantizing high-frequency components with a larger step size (i.e., more coarsely). This is because high spatial frequencies require less detailed coding. This step discards visually unimportant information and thus makes the approach "lossy." These quantized coefficients are then entropy encoded, with the DC coefficients being treated specially. Since the DC coefficients are a measure of the average value of the pixels in the block, they are expected to show less variation within the same component, and therefore, are differentially encoded.<br><br>The quantized AC coefficients are ordered in a zigzag sequence starting from the top-left corner and traversing the nearest cells first. This ordering puts the low-frequency coefficients before the high-frequency ones, and thus facilitates entropy coding. The coefficients are first run-length encoded and then coded using Huffman or arithmetic coding. <br><br>The output from the entropy encoder is the output of the JPEG encoder. The JPEG decoder simply reverses this process, using an entropy decoder, dequantizer, and the inverse DCT to reconstruct the image.<P><HR><h3><A NAME="hmode">The Hierarchical Mode of the JPEG Standard</A></h3>The hierarchical encoding mode of JPEG (HJPEG) encodes an image at multiple resolutions,each differing from its adjacent level by a factor of two in the horizontal or vertical directions, or both. The image is first bandsplit on the basis of spatial frequency and subsampled by the desired number of multiples of two in either or both dimensions. For sequential DCT-based HJPEG this new reduced size image (called a frame in JPEG terminology) is encoded using the sequential mode described previously. Then, the encoded reduced-size image is decoded, interpolated and upsampled by twohorizontally and/or vertically. This upsampled image is then used as a prediction of the original image at this resolution and the difference image is computed. The difference image (called a differential frame) is then encoded using the sequential mode. Finally, the last two steps are repeated until the original image at full resolution has been encoded.<P><HR><h3><A NAME="diff">Difference between Progressive Mode and Hierarchical Mode</A></h3>In the progressive mode, the 8x8 DCT coefficients blocks are encoded in multiple scans. Like the hierarchical mode, the progressive mode requires an image-sized buffer for storing the coefficients between quantizer and the entropy coder. Two procedures, <em>spectral selection </em> and <em>successive approximation</em>, are defined for encoding the quantized coefficients in thebuffer. Spectral selection encodes a specified band of coefficients in each scan. Successive approximation encodes the most significant bits in the first scan and then the less significant bits in each subsequent scan.<br><br>The hierarchical mode, as described previously<!WA5><A href="#ref2">[2]</A>, provides a progressive presentation with increasing spatial resolution between progressive stages. The image quality at extremely low bit rates is reported to be superior to any other modes of JPEG including the progressive mode<!WA6><A href="#ref3">[3]</A>.<br><br>Both the progressive and the hierarchical mode allow the progression of image, from coarse to refined. However, the multiple resolutions in the hierarchical mode provide an edge over the progressive mode in the following two cases. <UL><LI> <B>Heterogeneous end user devices</B>: When a user has a low resolution display device, a maximum resolution for that device is always possible without receiving unnecessary data for higher resolution if the hierarchical mode is used. Both the required bandwidth and the transmission time are reduced for a lower resolutiondisplay device. In the case of the progressive mode, a user can still see the progression of an image, however, the total transmission time is fixed for the full-resolution. Therefore, the progressive mode does not attend to various user devices efficiently. The extended use of network bandwidth (and transmission time if progressed to the full-resolution completely) incurs an unnecessary cost if the devices have lower resolutions than the coded image.</LI> <br><br><LI> <b>Multimedia data storage</b>: Thumbnail images are feasible in the hierarchical mode. However, the separation of an image by layer for storing in a different storage device is not justified in the progressive mode, since a progression up to full-resolutionmay be necessary regardless of image sizes.</LI></UL>Note also that the hierarchical mode can use other modes of JPEG for the progression. Thus, even a finer granularity of the progression is possible if the progressive mode is used instead of sequential mode. On the other hand, the progressive mode of JPEG does not have any overhead in bit rate over the sequentialmode of JPEG, whereas the hierarchical mode does.<br><br><DL><HR><h3><A NAME="refer">References</A></h3><A NAME="ref1">[1]</A> G.K. Wallace. "The JPEG Still Picture Compression Standard,"<em> Communications of the ACM</em>, 34(4):30--44, July 1991.<br><br><A NAME="ref2">[2]</A> J. K. Han and G. C. Polyzos, "Networking Applications of HierarchicalJPEG," <em> 1996 IEEE International Phoenix Conference on Computers andCommunications </em>, IPCCC'96, Phoenix, AZ, March 1996.<!WA7><A href="http://www-cse.ucsd.edu/users/jhan/conf/ipccc96.abs.txt">Plaintext Abstract</A>,<!WA8><A href="http://www-cse.ucsd.edu/users/jhan/conf/ipccc96.ps.gz">PostScript</A> (7 pages, 1.6M)<br>(also Technical Report CS95-425, Dept. of Computer Science andEngineering, University of California, San Diego, La Jolla, CA, May 1995)<P><A NAME="ref3">[3]</A> W.B. Pennebaker and J.L. Mitchell."JPEG Still Image Data Compression Standard," <em>Van Nostrand Reinhold</em>, 1993.<br><br><hr><DT>Last updated: December 12, 1995</DL><!WA9><A HREF="http://www-cse.ucsd.edu/users/jhan/hc.html">[Top]</A><!WA10><A HREF="http://www-cse.ucsd.edu/users/jhan/index.html">[Back to Home Page]</A></BODY><P></HTML>⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -