Multiple stage vector quantization using competitive learning

Abstract

The frequency sensitive competitive learning (FSCL) algorithm requires an excessive amount of training for vector quantizers with large codebooks. The authors present a possible solution to this problem through the application of the multiple stage vector quantization (MSVQ) technique to the FSCL vector quantizer for use in the coding of image data. By combining a number of smaller code books which can be trained quickly using the FSCL VQ, the MSVQ allows formation of a large effective codebook size that makes substantial reduction of the computational cost and improved performance possible. The MSVQ technique also allows implementation of a large codebook with a much smaller storage requirement than if the same size codebook were implemented using the basic FSCL algorithm.>

Related Papers

• → An Efficient Vector Quantization Codebook generation using a Triangle Inequality(2012)2 cited
• → Application of sorted codebook vector quantization to spectral coding of speech(2002)2 cited
• → Multiple stage vector quantization using competitive learning(2003)1 cited
• → Fast Vector Quantization Algorithm Based on Vector Features(2010)
• Comparative Study on the Application of LBG and SOFM in Image Vector Quantization(2010)