Variable rate vector quantization for medical image compression

Abstract

Three techniques for variable-rate vector quantizer design are applied to medical images. The first two are extensions of an algorithm for optimal pruning in tree-structured classification and regression due to Breiman et al. The code design algorithms find subtrees of a given tree-structured vector quantizer (TSVQ), each one optimal in that it has the lowest average distortion of all subtrees of the TSVQ with the same or lesser average rate. Since the resulting subtrees have variable depth, natural variable-rate coders result. The third technique is a joint optimization of a vector quantizer and a noiseless variable-rate code. This technique is relatively complex but it has the potential to yield the highest performance of all three techniques.

A progressive universal noiseless coder

“…In the first case, an additional (L+l) log ( N / L + l ) bits are required to specify the count and the reproduction vector (centroid) of the left child t L . This accounts for the remaining terms in (8). From this information, the decoder can derive the count and centroid of the right child t R by ntR = nt -ntL and w t R = (ntut-ntLwtL)/ntR.…”

“…First, the centroid of the entire data sequence is transmitted using L log ( N / L + l ) bits. This is the reproduction vector used at the root, and accounts for the first term in (8). With this initial tree TO, the decoder can already begin reconstructing the received image.…”

“…Since each component is quantized, the final image is lossy. More recently, pruned TSVQ [7], [8] has been used for progressive transmission. The image is scanned, and incremental descriptions corresponding to a sequence of nested, optimally pruned subtrees of a given TSVQ are sent to the decoder.…”

A competitive continuous Hopfield neural network for vector quantization in image compression

“…A great many vector-quantization approaches have been proposed for image compression by several researchersÐLinde et al (1980), Gray (1984), Chang et al (1988), Netravali (1988), Gersho and Gray (1992), Riskin et al (1990), Yair et al (1992), , and Andrew and Palaniswami (1996). In addition, neural-network-based techniques have been used to address vector quantization.…”

A Compensated Fuzzy Hopfield Neural Network for Codebook Design in Vector Quantization

“…A great deal of literature based on vector quantization has been discussed in other articles. 2,4,9,10,14,[17][18][19]24,28 The goal of vector quantization is to create a codebook for which the average distortion generated by a training vector and a codevector in codebook is minimized. The minimization of average distortion is widely used by a gradient descent-based iterative procedure.…”