Wavelet Image Coding Algorithm Based On Human Visual

Mallat presents the idea of multi-resolution analysis, and units the constructed way of all the wavelet functions in 1989, so wavelet transform (WT) is applied widely in image compression. Following development of image coding based on WT, many classic methods of image coding has been proposed. Among them, Said and Pearlman's Set Partitioning in Hierarchical Trees (SPIHT) is well known for its simplicity and efficiency as one of algorithms based on wavelet transform and zerotree quantization. Based on zerotree structure, it is an efficient embedded image compression algorithm, utilizing characteristics of strong correlation among wavelet coefficients of different resolutions in the same direction. But the algorithm is memory consumed and not easy to implement on hardware. The paper describers the math theory base of WT and the characteristics of WT coefficients, improves the algorithm utilizing human visual characteristics to achieve better performance of the algorithm at low bit rate.Firstly, it introduces continuous and discrete wavelet transform according to analysis of the disadvantage of Short Time Fourier Transform. From the viewpoint of multi-resolution analysis, Mallat algorithm is deduced.Secondly, the character of the wavelet transform used in image compressions is researched, including the characteristics of WT coefficients, and how to assess the reconstructed image after wavelet image compression.Finally, still image compression algorithm is proposed for grayscale image and color image respectively in DWT domain. The coding algorithm to grayscale image is implemented by weighting the wavelet coefficients according to the contrast sensitivity masking; The coding algorithm to color image considers human's sensitivity to different components of color image and strong correlation of different components in image space. It generates corresponding model to improve SPIHT respectively. The advantage of the improved algorithm reflects in two terms: it gives different visual weights to wavelet coefficients in different subbands of the different components after DWT, making sure of priority transmission to the coefficients human visual are sensitive to advance the quality of the reconstructed; it sufficiently makes use of the distribution character of wavelet coefficients and decreases redundancy of bitstream to boost efficiency of the coding algorithm. Experimental results show that the proposed technique improves the quality of the reconstructed image in both PSNR and perceptual result, compared to SPIHT at the same bit rate.