An error-free image compression algorithm using classifying-sequencing techniques.

Digital image compression is more and more in demand as our society becomes more information oriented, with more digital images being acquired, transmitted and stored everyday. Error-free, or non-destructive, image compression is required in applications where the final image is to be analyzed digitally by computers. A new error-free digital image compression algorithm, the Classifying-Sequencing algorithm, is presented in this dissertation. Without the help of any statistics information of the images being processed, this algorithm achieves average bits-per-pixel close to the entropy of the neighboring pixel difference. In other words, the compression results are comparable to the best that a statistics code can achieve. Because this algorithm does not involve statistical modeling, generation of a code book, or long integer/floating point arithmetics, it is simpler and therefore faster than the standard statistics codes, such as Huffman Code or Arithmetic Code. In this dissertation the new algorithm under discussion is tested using seven images, together with several known algorithms. Three lower-order entropies of the image files are also provided for comparisons. Presenting compression results from an isolated algorithm is not sufficiently objective for comparisons between algorithms, as potential discrepancies exist between not only different images but also same images when reproduced from prints. Comparing the results of different algorithms and with the entropy of the neighboring pixel differences on the same images is more objective. When the entropy of an image is high, the compression ratios of all algorithms are likely to be low; and vice versa. Given that it is faster in decoding than in encoding images, the most prospective applications of the Classifying-Sequencing algorithm are in the fields of digital image transmission, distribution and archiving, where the images are likely to be encoded once but decoded many times. It can be easily realized on simple processors, or completely in hardware, due to its simplicity