Bezztrátová komprese JPEG grafiky

• Main Author: Ondruš, Jan
• Other Authors: Žemlička, Michal
• Format: Dissertation
• Language: Czech
• Published: 2009
• Online Access: http://www.nusl.cz/ntk/nusl-286256

JPEG is a commonly used method of compression for photographic images. It consists of lossy and lossless part. Static Huffman coding is a last step. We can replace this step using advanced techniques and arithmetic coding. In this work we introduce method used for additional compression JPEG files (files in JFIF format) saved in baseline mode. Partial decompression is a general way we can use in this case. We invert only last lossless steps of JPEG compression algorithm. Compressed file is transformed into array of quantized DCT coefficients. We designed algorithm for prediction of the DCT coefficients. It returns particular linear combination of previous coded coefficients in current and neighbouring blocks for each from 64 coefficients in block matrix. We show how this prediction can improve efficiency of compression of JPEG files using Context Mixing algorithm implemented in PAQ8 by Matt Mahoney. Specific implementation is described and its compression ratio is compared with existing methods and applications for further lossless JPEG images compression.