Segmentation-based vector quantization of images by a competitive learning neural network

The authors present a segmentation-based VQ technique using a competitive learning neural network, which significantly improves the preservation of edge characteristics and greatly reduces the computational complexity and memory requirement. Unlike most segmentation-based techniques, an adaptive image segmentation method has been developed and is used to segment edges from images without the need of any preset thresholds. Experimental results show that the reconstructed images have no perceptibly ragged edge effect. Compared with results from other segmentation-based block coding techniques, the method achieves better performance at a lower bit rate (or a higher compression ratio)