A Study on Image Compression Based on Vector Quantization Technique

• Main Authors: A-Pei Tsai, 蔡阿佩
• Other Authors: Ju-Yean Hsiao
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/57004414081621779171

碩士 === 國立彰化師範大學 === 資訊管理學系 === 90 === Internet becomes a popular transmission channel recently. Most data such as text, image, audio and video can be represented in digital form and transmit on the network. As we know the bandwidth is restricted, too many multimedia data transmit over Internet result in congestion or delay. The performance of network is degraded. In order to increase the compression rate but preserve more image detail, many image compression methods are proposed in the last decades. Since the human visual system is low sensitivity, we can reduce the image size but still keeping the image quality in allowable range. In this thesis, two methods based on Vector Quantization (VQ) for lossy image compression are proposed. In the first scheme, we add an extra bit per block to indicate whether the block needs to search the nearest codeword or not. Each block is compressed according to the variance of it. If the variance is less than the predefined threshold, we represented the pixel values in the block with the mean of the block. Otherwise, we choose a nearest codeword to represent the block. We have less calculating time and better image quality than VQ after reconstructed. The experimental results reveal the compression rate of our method is almost as good as VQ. In the second scheme, we propose a post-processing method for VQ compressed image. We apply lossless encoding technique, predictor and the concept of state codebook for lossless image compression. In our experimental results, the bit rate is lower than previous methods. We discuss the size of state codebook further. For the variable size approach, we set two thresholds to decide the size of state codebook. According to the experimental results we can find that the compression ratio of variable size is better than fixed size.