基于图像低频子带极大值映射的量化算法  

Quantization algorithm based on images lowpass subband maxima mapping

作　　者：黄胜[1] 杜呈尘 翦伟 

机构地区：[1]重庆邮电大学光通信与网络重点实验室,重庆400065

出　　处：《计算机应用》2015年第11期3288-3292,共5页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61371096;61171158;61275077);重庆市自然科学基金资助项目(cstc2013jcyj A40052;cstc2012jj A40060);重庆市教委科学技术研究项目(KJ130515)

摘　　要：针对图像压缩中的死区量化不能有效保留图像边缘信息的问题,提出了低频子带极大值映射量化算法。在图像经过小波变换后所形成的各级子带中,首先利用与低频子带系数呈映射关系的各级高频子带系数的均值确定低频子带中各系数的重要性。在量化过程中,高频子带系数采用JPEG2000中的死区量化步长进行量化,低频子带系数根据自身重要性自动更新量化步长,从而有效保留图像的边缘信息。提出的算法在量化步长更新时对低频系数的选择具有自适应性的优点,与传统的JPEG2000算法相比,所提算法能够加快优化截断的嵌入式分块编码(EBCOT)阶段Tier1的编码速度。实验结果表明,所得图像证明了此算法在保留图像的边缘信息方面具有一些优势,所提算法的峰值信噪比与传统的死区量化相比有约0.2 d B的提升。Concerning the problem that deadzone quantization in image compression cannot protect the edges of images effectively, a novel quantization algorithm called Lowpass subband Maxima Mapping Quantization （LMMQ） was proposed. In all kinds of subhands after wavelet transform, the importance of the coefficients in lowpass subbands could be decided by the average value of all coefficients in highpass subbands which have a mapping relationship with the coefficients in lowpass subbands. During quantization, the coefficients of highpass subbands were quantized by deadzone quantization in JPEG2000. The quantization step size of coefficients in lowpass subbands could be adaptively refined because of their own importance, so the edges of images could be protected effectively. The proposed algorithm has an advantage of adaptability in the aspect of coefficient selection when the step size is refined, and has higher encoding speed in Tier1 of EBCOT （ Embedded Block Coding with Optimized Truncation） than traditional JPEG2000. The experimental results show that the proposed algorithm has an advantage of protecting the edges of images and has 0.2 dB more than traditional deadzone quantization.

关 键 词：图像压缩 小波变换 边缘保留 自适应 量化步长更新 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]