In this thesis the performance of binary arithmetic coding(BAC) is evaluated in image data compaction.
BAC is a data compaction technique that encodes the binary symbol string by creating a code string which represents a fractional value on the number line between 0 and 1. A k-ary symbol string can be encoded with BAC when it is converted to a binary symbol string, which can be compacted by classification and statistics bins.
Four methods are compared in a DCT based compression system with four test images. They are the Huffman coding and the BAC with three different conversion methods in k-ary to binary conversion. : simple binary representation, binarization by tree-decision, tree-decision and classification. In order to see the coding efficiency according to the changes of statistical characteristics, quantization scaling factors(SF) are changed from 1 to 4.
Experimental result shows that the BAC method with tree-decision and classification has the best coding efficiency, and has better efficiency than Huffman coding based on the ISO/CCITT code table by about 4% when SF is 1. This BAC is also more robust than Huffman coding in the changes of statistical characteristics.