Data Compression Standards

This paper first explores the history of the Digital Video MPEG Standards as created by the MPEG Committee of the International Standards Organization(ISO) for data compression. Included are existing standards, i.e. MPEG-1,2,4, and 7. Additionaly, recent developments to the MPEG-21 standard that is still in production will be looked at. For each of the standards, a brief overview of the standard will be given, followed by a summary of some of their key points, technologies, and abilities. The paper then uses this key information to discuss data compression of different media including video, audio and multi-media applications, Examples will be given to show differences between algorithms used for the data compression. Each section of the paper will cover key concepts and ideas to provide the reader with a suitable background for further research. Introduction By introducing powerful standards for the compression, representation, and organization of digital video, the MPEG group, a division of the ISO, made it possible for the widespread use of digital video. Currently, there are five Digital Video Standards that the MPEG committee has introduced, from MPEG-1 to MPEG-21, which is still in its early stages. The format of the paper is as follows: 1. MPEG-1 (A standard used primarily in the storage of video) 2. MPEG-2 (The standard preferred by DVD, SVCDs) 3. MPEG-4 (Designed with multimedia applications in mind) 4. MPEG-7 (Describes all the different facets of a video) 5. MPEG-21 (Creates a multimedia framework) For each of the standards, a brief overview of the standard will be given, followed by a summary of some of the key points, technologies, and abilities of each standard. It’s impossible to go over each of the standards in great detail, but an effort will be made to provide the reader with a basic understanding of the material. DATA COMPRESSION STANDARDS 2 Figure 1. The lack of change from one frame to the next MPEG-1 Standardized in 1991, MPEG-1 is the oldest of the digital video standards produced by the MPEG group. The purpose of MPEG-1 was to provide technologies and algorithms for the efficient compression of digital video. This was done to facilitate the transport and storage of digital content, while maintaining quality. Video content is very redundant because it is made up of a series of still frames. As such, the difference between frames in a given sequence is minimal, if it changes at all. There may also exist redundancies within each frame, i.e. a certain color could be used to represent an entire area of pixels. Figure 1 shows two sequential frames that the author ripped from a DivX video. It shows the lack of change from one frame to the next. The algorithms and technologies of MPEG-1 are designed to take advantage of these redundant elements. The importance of the techologies and concepts defined by MPEG-1 for the compression and transport of video is that they form the basis for later standards. To cement the basic concepts of MPEG-1, a real-life example of how compression works will be demonstrated, followed by a look at the encoder-decoder block algorithm. MPEG divides videos into sequences that group together similar frames. A frame is nothing more than a still image. The frames, in each sequence, are then divided into groups of I, P, and B-Frames (explained below). These frames are important in determining how compressed a frame can be, the chance of error propagation, whether it represents interframe or intra-frame data, and the encoding order. If a frame is an I-frame, then it is an intra-coded frame. Being intra-coded means that it looks for redundacies only within its own frame. Because it is self-referencing, I-frames are always used to start a video-sequence. Also called the ”key-frame,” an I-frame is the frame that media players point to when moving through a videoSikora, T.Ebrahimi and M.Kunt. DATA COMPRESSION STANDARDS 3 The P-Frame (Predictive) uses information found in previous frames and then performs motion compensation (MC) to ”guess” what its values are. P-frames achieve a higher degree of compression compared to I-Frames, but can be risky since they can be used to reference other P-Frames, which may cause errors to propagate. B-Frames (Bi-directional) are like the P-Frames, but look at future frames as well as previous frames to construct themselves. The B-Frame achieves even higher compression than a P-Frame and doesn’t have propagation errors because it’s not used as a reference for any other frame. These 3 frames are then grouped together in many different ways to form a video sequence, which starts off with an I-Frame, and then includes a series of more P, B, and I-frames. Thus sequences like IPPBPP or IPPIIBIIP are valid sequences of frames. Of course, there are some that provide better compression than others, but many sequences are possible. Once the frame type ratio and ordering is worked out, the video can be encoded. To organize a frame for compression, MPEG-1 follows a divide and conquer approach. Individual frames are cordoned off into a grid of squares called macroblocks. The standard size for intra-coded macro blocks is 8x8 pixels, where each pixel represents a single color. Each of 64 pixels in a macroblock are encoded and compressed and then reconstructed by the decoding process (Watson, 1994; Sikora, 1997, ; T.Ebrahimi & M.Kunt, 1998). There are three steps in standard MPEG video compression 1. The image is transformed into a simpler representation. 2. The numerical values from step 1 are quantized. 3. Finally the quantized values from step 2 are encoded using entropy. If done properly, and the encoded video is smaller than the original, then compression has occured. To make the process clearer, lets go over each of the steps in some detail. The first step in the compression process is to transform the frame’s macros into something simpler and smaller. There are a variety for techniques that do this; the one commonly used by MPEG is the Discrete Cosine Transform(DCT) (Sikora, 1997, ). The DCT basically takes a matrix of values representing a macroblock, an 8x8 macro in this case and, using various vectors and cosines, transforms the pixel values into something smaller (Watson, 1994). There are a few different DCT transforms used in image compression. The one used in this paper is as follows: For any size matrix nxm, the value of the transformed pixel at point (u, v) is given by the equation in Table 1. Table 1: DCT Transformation Equation S(u, v) = 2 √ nm C(u)C(v) m−1