基于空间累进预测的图像压缩方法-智能与分布计算实验室

基于空间累进预测的图像压缩方法

姓名	许铁军
论文答辩日期	2005.05.11
论文提交日期	2005.05.14
论文级别	硕士
中文题名	基于空间累进预测的图像压缩方法
英文题名	An Image Compression Scheme Based on Spatial Progressive Prediction
导师1	卢正鼎
导师2
中文关键词	预测编码;空间累进;嵌入式零树小波编码;多级阈值量化
英文关键词	Predictive coding, Spatial progression, Embedded zerotree wavelet, Multilevel thresholds quantization
中文文摘	作为传递信息的媒体和手段之一,图像信息十分重要。但广泛使用数字图像还有一个主要的障碍，就是数字图像的数据量非常大，需要很大的空间来存储，因而需要对图像进行压缩。近年来，人们对静态图像的压缩进行了深入研究，提出了一些非常好的方法，如基于离散余弦变换（DCT）的JPEG和基于离散小波变换（DWT）的JPEG2000，特别是DWT已经成为了图像压缩中最重要的核心技术之一。虽然这些方法效果非常好，但是都存在共同的缺点，就是计算量比较大，同时对内存要求也比较高。该缺点限制了这些方法在计算能力较弱的环境中的应用。传统的空间预测编码方法虽然计算简单，但是由于只能利用光栅扫描顺序中前面的像素对当前像素进行预测，因此效果不很理想。针对基于变换的和基于预测的压缩算法的缺点，结合二者的优点，提出一种结合空间累进的预测编码，在预测过程中能够利用被预测像素四周的像素对该像素进行预测。通过构建层次结构图来将图像中的像素划分等级，根据层次等级不同依次进行预测，预测过程中采用就地处理的思想，以减少对内存的需求。根据不同等级像素重要性不同的特性，采用改进的多级阈值嵌入式零树小波编码方法对预测结果进行多级阈值量化。实验结果表明，此方法计算简单，内存要求小，性能优于JPEG，在实时和无线多媒体等计算能力较弱的环境中尤为适用。
英文文摘	As one of the method of transferring information, image is very important. However, the predominant drawback to apply digital image is the bulk data that needs much space. In order to overcome this drawback, we must compress the digital images. Static image compression has been deeply researched these years, and some excellent schemas were found, such as the JPEG standard based on discrete cosine transform(DCT) and the JPEG2000 standard based on discrete wavelet transform(DWT).The result of these schemas are satisfied. In particular, DWT has even become one of the most important core technologies of image compression. However schemas using transform have a common defect, which is the computational complexity and more memory required. Therefore, this may restrict the applications of them in the environments of low-computational power，such as real-time and wireless multimedia. Traditional spatial prediction based image compression approaches have lower computational complexity. But because such approaches use only preceding pixels according to the order of the raster scan as the input to predict the sample pixels, their results are not satisfied. Considering the defects of both transform based and prediction based schemas, and integrating their values, a new image compression scheme of progressively spatial prediction is proposed, pixels around the predicted pixel can all be used. Pixels of an image are divided into different levels, and predicted according to their levels. In-place processing is used in the process of prediction to lessen the requirement of memory. Considering pixels of different levels having different significance, we utilize improved embedded zerotree wavelet (EZW) to get multilevel thresholds quantization. Experimental simulations show that our scheme have following features: simple computation, less memory requirement, and better performance than JPEG. It is especially applicable to limited computation power environments such as real-time and wireless multimedia.