Real-Time Aggressive Image Data Compression

Abstract : The objective of the proposed research is to develop reliable algorithms that can achieve aggressive image data compression (with a compression ratio of 60 times or more) for real-time implementation. Typical applications of such algorithms include terrestrial HDTV broadcasting, space communications, and handling and disposing of toxic materials and nuclear wastes with remotely controlled robots. The state-of-the-art techniques are hampered by serious technical barriers of codebook design complexity. The proposed approach is built on a vector quantization (VQ) algorithm recently developed by the PI. The codebook design complexity of this VQ algorithm is only linearly proportional to the codebook size (significantly less than conventional algorithms) and the encoding complexity is independent of codebook size. Highlighting the proposed approach is a piecewise-linear transform preceding VQ based on the concept of entropy partitioning. The novelty of the proposed algorithm is due to the following: (1) introduction of a piecewise-linear transform to VQ so as to retain more input information: (2) exploiting both inter-block and intra-block redundancy, (3) use of parallel distributed network for real-time codebook design. The proposed research is significant as (1) it addresses the imminent demands of solving the aforementioned real-world problems; (2) its accomplishment will alleviate the serious complexity barrier of conventional VQ algorithms; (3) it pushes forward the technical frontiers of data compression.