Hybrid Image Compression Algorithm Based Wavelet And Contourlet

As the key technique of modern communication, media storage, data publish and multimedia computer, image compression is one of the most active areas in information processing. The digital image has huge data information, so the algorithm which has high compression ratio and high fidelity rate is urgently needed, taking account of storage cost and communication.Nowadays, wavelet-based image coding technique is the main trend in image compression field. However, wavelets can not efficiently represent and process multimensional information. Contourlet transform is an efficient anisotropy analysis tool in high dimensions. Compared to wavelets, contourlets not only possess the main features of directionality and anisotropy, but also can effectively capture the intrinsic geometrical structures such as smooth contours in natural images.In this paper, firstly, the classic image coding algorithms in wavelet domain are introduced, and the contourlet theory and the state of art of its application to image coding are analyzed. Secondly, three new image coding algorithms are proposed, which are the main point of this paper. The new low bit-rate image coding algorithm in DWT domain, which using the statistical characteristics of the lowest frequency coefficients and the correlations between coefficients in highpass bands, optimizes the lowpass band and proposes adaptive scan orders of highpass bands to complete image coding; The new image coding algorithm based on DWT and Coutourlet, the wavelet transform is applied to original image, and adaptive directional decompositions are employed to middle and high frequency subbands. Then, the lowest frequency coefficients are optimized, a new extended spatial orientation tree is constructed, and the encoding scheme is realized by the idea of SPIHT; The new vector quantization image coding algorithm in wavelet and coutourlet domain. The hybrid decomposition is applied to original image at first. Then, the lowest frequency coefficients are encoded by DPCM. And, the vectors are constructed with adjacent coefficients from the same band at the same location. The codebook are trained by competitive learning vector quantization, which using a new initialization codebook design. Finally, the outputs are encoded by Huffman codec. At the end, the three algorithms are concluded, theoretical and practical significance of this paper is obviously.Experimental results show that all the three proposed algorithms improve the quality of the reconstructed image when compare to classic or advanced schemes. At low bit rate, our algorithms possesses the features of interoperability and adaptability, which has excellent performance in both PSNR and perceptual results.