This study is divided into two parts. The first part involves an investigation of near-lossless compression of digitized images using the entropy-coded DPCM method with a large number of quantization levels. Through the investigation, a new scheme that combines both lossy and lossless DPCM methods into a common framework is developed. This new scheme uses known results on the design of predictors and quantizers that incorporate properties of human visual perception. In order to enhance the compression performance of the scheme, an adaptively generated source model with multiple contexts is employed for the coding of the quantized prediction errors, rather than a memoryless model as in the conventional DPCM method. Experiments show that the scheme can provide compression in the range from 4 to 11 with a peak SNR of about 50 dB for 8-bit medical images. Also, the use of multiple contexts is found to improve compression performance by about 25% to 35%;The second part of the study is devoted to the problem of lossy image compression using tree-structured vector quantization. As a result of the study, a new design method for codebook generation is developed together with four different implementation algorithms. In the new method, an unbalanced tree-structured vector codebook is designed in a greedy fashion under the constraint of rate-distortion trade-off which can then be used to implement a variable-rate compression system. From experiments, it is found that the new method can achieve a very good rate-distortion performance while being computationally efficient. Also, due to the tree-structure of the codebook, the new method is amenable to progressive transmission applications.