Large Image Compression Algorithm Research For Hardware Implementation

The theory of discrete wavelet transform (DWT) is presented in 1970's and fast developed from 1980's. The coding efficiency and reconstructed image quality are better using DWT than using the traditional DCT. Therefore, it has been adopted by JPEG2000 standard. The aim of this thesis is researching wavelet-based compression for large image, improving and creating algorithms to suit hardware implementation.Spatial combinative lifting algorithm (SCLA) is proposed under the scheme of lifting. It combines the horizontal and vertical transforms of 2D-DWT, resulting in a great decrease of multiplication amount. Due to restriction of memory storage in hardware, a block-based SCLA is designed, which does not increase the number of computing, but makes the transform buffer not dependent on the image width.According to the probability distribution function of high frequency subband coefficients, the best quantization steps of uniform quantization are calculated for different number of floors, as well as the outputting entropy. The expected code rate can be assigned into different subbands correlated with their variances, and then the quantization parameters can be determined by the relationship between entropy and quantization steps. Scanning direction of run-length coding is also modified to fit the spatial structure of high frequency coefficients.Tree-based coding algorithm is a family of coding methods for wavelet coefficients. It represents the multi-resolution characteristic of wavelet decomposition so that it can obtain better performance than traditional methods. Some classical tree-based algorithms are discussed in this thesis and modified for hardware implementation.The research leads to the result that a good fast wavelet-based compression algorithm for large image is feasible in a memory-constrained hardware environment.