Classification, compression and transmission of chromosome images for genomic telemedicine

In this dissertation, we address three fundamental issues in genomic telemedicine systems: chromosome classification, chromosome image compression and chromosome image transmission. We first treat chromosome classification as an appearance-based object recognition problem. Optimal linear subspaces, derived from the linear discriminant analysis and principal component analysis, are studied with respect to Bayes and nearest-neighbor classifiers. Our study suggests that classifier dependency is an important but previously unaddressed issue that should be taken into consideration in the pursuit for an optimal subspace. We then propose a cascaded differential and wavelet-based scheme for chromosome image compression. The compression algorithm combines lossless compression of chromosome regions of interest with lossy-to-lossless coding of the remaining image parts, instead of compressing the entire chromosome image as a whole in current commercial systems. Finally, we propose a unified framework for addressing progressive chromosome image transmission over noisy channels based on the finite-state Markov channel (FSMC) model. Using a concatenation of rate-compatible puncturing convolutional code and cyclic redundancy code for error protection, we analytically derive optimal joint source-channel coding solutions in the form of unequal error protection (UEP) for chromosome image transmission over FSMCs. Fast algorithms are proposed to search for the optimal UEP solutions for real-time applications.