Iterative hillclimbing optimization techniques for transform image encoding/decoding and for image segmentation

An iterative hillclimbing optimization technique was introduced in this thesis. It was used to tackle many index assignment problems, i.e. transform image encoding, noisy image decoding; and image segmentation. We first studied basic transform image coding techniques, then introduced an iterative algorithm which has a hillclimbing property on the cost function. We then extended the algorithm to hyperspectral image coding. We realized that the algorithm can be generalized to other applications as well. We applied the iterative hillclimbing idea to noisy channel image decoding. We also investigated a Turbo-like joint source-channel decoding technique, which is another kind of iterative decoding. Lastly, we re-investigated the image segmentation application, using the iterative hillclimbing algorithm. The hillclimbing method inspired development of another iterative algorithm which is an extension to mean-field annealing, with applications both to image decoding and image segmentation. The hillclimbing algorithm at the heart of this thesis thus yielded several promising offshoot directions for continuing research.