| This thesis describes fault-tolerant designs of JPEG and JPEG2000 image compression systems. Fault models are proposed for each compression stage based on data types and numerical algorithms implemented in the stage. Concurrent error detection and error control coding techniques are used to protect the compression systems from errors. The work starts by developing fault models for each individual stage of the image compression processes, such as data transformation, discrete cosine transform, wavelet transform, quantization, entropy coding, embedded block coding, and JPEG-JPEG2000 compression formats. Theory and practice of setting numerical tolerances are discussed to separate errors caused by the faults from inherent finite-precision arithmetic calculations. The designs are evaluated by computer simulations which measure the error recovery performance. |
