The HTVQ model can be created in the VisualSim schematic editor called ModelBuilder, as shown below:
Each frame is first subdivided into blocks of 4 pixels by 2 pixels. Each block is then passed through a Vector Quantizer, which selects the best reconstruction block from a precomputed codebook. The output of the Vector Quantizer is the codebook index of the reconstruction vector. Each block is decompressed by indexing into the codebook, and then reassembled into the output image.
The Vector Quantizer in this implementation does not actually find the optimal reconstruction vector for each block. Instead, the quantization is done using a series of 3 table lookups. The resulting quantizer is very close to optimal and results in a significantly faster implementation. This technique is known as Hierarchical Table-Lookup Vector Quantization (HTVQ).
This model uses a custom applet to achieve a layout different from that provided by default.
Copyright 2005 © Mirabilis Design Inc. and University of California Berkeley. All Rights Reserved.