Optimization and representation techniques for multimedia processing software.

摘要

Flavor, Formal Language for Audio-Visual Object Representation, has been introduced to simplify and speed up development of software that processes coded multimedia or any bitstream-oriented data. It is a language for describing bitstream syntax, and from the description, its translator can automatically generate C++ or Java code that can be used for bitstream reading, writing, and tracing.; A Flavor description only specifies how the data is laid out on the bitstream, and as a result of this abstraction, efficient algorithms can be easily applied for automatic parsing and generation of bitstreams. The crux of the multimedia bitstream parsing and generation deals with compression, where the most widely used statistical coding methods are variable-length coding, e.g., Huffman coding, and binary arithmetic coding. "Customizable" methods for efficiently processing such coded data have been introduced, and the Flavor translator has been enhanced to include them in the generated code. Additionally, algorithms that automatically customize the methods have been added to the translator so that the methods work efficiently for each given set of codewords or each given binary arithmetic coder description. We show that our automatically generated methods are faster than or comparable to other known, manually designed methods.; As a complementary solution to Flavor and its translator, we present XFlavor, a framework for providing XML representation of multimedia data. XFlavor can be used to convert multimedia data back and forth between binary and XML representations. Compared to bitstreams, XML documents are easier to access and manipulate, and consequently, the development of multimedia processing software is greatly facilitated, as one generic XML parser can be used to read and write different types of data in XML form. On the other hand, XML representation inflates the original binary data, and for storage or transmission, a more compact form is required. For this, the original bitstream syntax is used to convert XML documents into much smaller binary data. We show that such conversion yields better compression than any one of the currently known XML compression tools.