Image Compression Schemes Based On Wavelet And Application To Video Compression

Image compression plays an important role in transmission and storage of multimedia. Wavelet has been widely adopted in image compression due to its multiresolution, time/frequency localization, lower time complexity, etc. With the advance of image compression techniques based on wavelet, a variety wavelet coefficient models are presented, such as EZW, SPIHT and EBCOT used in JPEG2000. They all are SNR scalable, and, especially EBCOT is resolution scalable. These properties are essential in transmission of image on Internet.Because the performance of an entropy coder can be significantly improved by having the coder dynamically adapt to the current "context", context models also have widely been adopted in image compression. Using context model, JPEG-LS and CALIC can improve their compression rate and gain higher lossless compression rate than EZW, SPIHT and EBCOT. Inspired by Glicbawls, CALIC, ECECOW, and EZW coding schemes, a new context model of wavelet coefficients for image compression is proposed. Wavelet coefficients are encoded by the arithmetic encoder, with the contexts being formed by quantizing linear prediction values. Experimental results show that the model achieves higher lossless compression rate of image than lossless SPIHT and lossless EBCOT used in JPEG2000. In addition, by exploiting the multiresolution property of wavelet, the model can deal with the transformed image for transmission purposes, which is resolution scalable, and earn higher compression rate for each scale of the image than EBCOT.Some coefficient models of wavelet for image compression have already been extended to video cases using Three-Dimensional wavelet. In the same way, after modification, PCW is also extended to 3D-PCW which is resolution scalable in time and space. First, the coefficients of wavelet are quantized by the threshold formed by the eye's sensitivity to it's subband. Then, Wavelet coefficients are encoded by the entropy encoder, with the contexts formed by quantizing their prediction values, which are determined by their neighbors in space and time dimensions. Since 3D-PCW tackles the coefficients as a whole, the context of a coefficient is computed only once, and the coefficient is encoded only once. But in 3D-ESCOT the context has to be computed once and the coefficent has to be encoded once on every bit plane. Therefore, 3D-PCW has much lower time complexity than 3D-ESCOT. Experimental results also show that for every quantization width, every scale in time and every scale in space, 3D-PCW not only achieves higher compression rate but also costs less time than 3D-ESCOT.