Wavelet-based Image Compression And Its Application To Face Recognition

With the development of multimedia technology, information of image plays a more and more important role in our life and at all times, people pays much attention to the technique development of image compression. As the merits of multi resolution analysis, wavelet theory now is widely applied in every fields of signal processing, especially in image codec, like JPEG2000, MPEG4 standards, etc. Wavelet based SPIHT (set partitioning in hierarchical trees) codec algorithm and its application in face recognition are studied in this thesis. The main contributions are as follows:1. The followings are studied in this thesis: principle of wavelet based image coding methods, key problems of wavelet transform in image codec and some main coding algorithms based on wavelet. The performance of SPIHT codec and JPEG(Joint Photographic coding Experts Group) codec is compared here, experiment results indicate that performance of SPIHT is better than JPEG's, no matter on objective or subjective assessment of image.2. The compression of noise image based SPIHT is realized. Such image can be de-noised while compressing. Result shows that de-noise ability to image with small noise is higher than those with large ones in the same case.3. Lossy and lossless image compression is realized by means of integer wavelet transform, and the shortcoming of pure lossy compression with non integer ones is overcame. Here the influence of different wavelet base to the effect of compression is discussed, and conclusions are obtained as follows: To the choice of wavelet base in SPIHT image codec, in lossless codec, integer LG5/3 and in lossy codec, non integer DB9/7 wavelet base is recommended.4. The influence of IAM (Image Activity Measure) to the extent of compression is discussed. Here image compression with preprocessing is studied and improved, the simulation results show that the preprocessing methods presented in thesis is much better than original ways.5. Based on SPIHT algorithm, a codec software packet is developed and used in face recognition system. Simulation results indicate that the recognition ration based compressed face images of 40:1 and 100:1 is the same as non compressed ones, and to the images of 200:1 compressed, the correctness is above 95%.