| 摘要 |
PROBLEM TO BE SOLVED: To solve the problems in computational quantity in vector quantization, using distorted encoding by a combination of companding functions and product codes. SOLUTION: A converterψ11 generates a feature vector y=[y<SB>1</SB>, ..., y<SB>m</SB>] by a functionψfrom an information source of an n-th dimensional input vector. Quantizers 12-1 to 12-m generate compression vectors y=[y^<SB>1</SB>, ..., y^<SB>m</SB>], obtained by making the components y<SB>1</SB>to y<SB>m</SB>of the feature vectors discrete by scalar quantum functionsΓ<SB>l</SB>toΓ<SB>m</SB>, respectively. An inverterϕ13 generates a decoding vector x^ by the functionϕ, from the compression vector y^. The transducerψ11 and the inverterϕ13 are each parameterized and are adaptively so adjusted that the best compression performance can be exhibited. Also, the quantization numbers of the quantizers 12-1 to 12-m are optimized. COPYRIGHT: (C)2005,JPO&NCIPI
|
