Low Complexity Implementation of Entropy-Constrained Scalar Quantization for Image Compression
Jianyu Lin · 2018
A low complexity implementation of entropy-constrained scalar quantization is introduced for coding the transformed coefficients in image compression. The Laplacian distribution is used to model the local magnitude distribution of the transformed subband coefficients. The entropy-constrained scalar quantizer for a Laplacian distribution is the uniform quantizer with a dead-zone at “0”. The sparsity of the quantized coefficients is found and used to determine the Laplacian distribution. Then, the threshold of the dead-zone is determined and used to trim the quantized coefficients. Since the Laplacian distribution is determined by the sparsity, the entropy-constrained scalar quantizer is implemented by a table looking process, making the complexity low. Coding results show that the introduced quantizer achieves higher compression efficiency than the double dead-zone scalar quantizer, which is commonly used in high-efficiency image compression algorithms.