一种基于分类矢量量化器的小目标红外图像压缩方法

A Small Target Infrared Image Compression Scheme Based on CVQ

本文提出了一种基于分类矢量量化器的小目标红外图像的压缩方法。首先利用图像子块的平均灰度与纹理能量这两个参数将图像划分为背景区域与感兴趣区域,然后分别对两类区域的子块进行码书设计,用相对较多的码字描述感兴趣区域,用相对较少的码字描述背景区域,这样既达到了较高的压缩比,同时又较好的保留了感兴趣区域的信息,并且编码计算量有大幅度的下降。文中对分类矢量量化器对于减小编码计算量的作用进行了理论分析。实验结果表明在相同码书尺寸的情况下本算法比直接矢量量化方法更好地保留了红外图像中的小目标信息,并且加快了编码速度。

This paper presents a small target infrared image compression scheme based on classified vector quantization. Firstly mean gray value and texture energy are employed as two features to classify all blocks into two types：regions of interest and regions of background. Then sub-codebooks are designed respectively according to the classification, so that bits are allocated preferentially to preserve those spatial regions in the image that have the high probability of being targets. Thus high compression ratios are reached while preserving information of interest. This paper also analyses the reduction of encoding computation of classified VQ comparing with direct YQ. Experiment results show that this compression scheme can preserve the small target information in infrared image better than direct VQ,and remarkably speed up the encoding process.