基于照射_反射模型和有界运算的多谱段图像增强  

Multispectral image enhancement based on irradiation-reflection model and bounded operation

作　　者：毕国玲[1,2] 续志军[1] 赵建[1] 孙强[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,长春130033 [2]中国科学院大学,北京100049

出　　处：《物理学报》2015年第10期74-82,共9页Acta Physica Sinica

基　　金：国家自然科学基金(批准号:60977001)资助的课题~~

摘　　要：根据多尺度照射_反射模型,结合广义有界运算模型和引导滤波,能够有效地解决多谱段降质图像的增强问题.算法采用自适应的引导滤波核函数作为环绕函数,估计反映图像整体结构的不同尺度的低频照射分量;利用有界广义对数比(general log-radio,GLR)模型加法代替Retinex理论中的对数变换运算;再由GLR模型减法去除照射分量,将不同尺度的反射分量从原始图像中分割出来;对不同尺度反射分量的有效信息采用有界GLR模型乘法和加法进行融合,有效地避免光晕伪影现象及越界现象的发生,得到多尺度反射分量图像,即最终的增强图像.通过对可见光波段的低照度图像和雾霾图像、红外图像、X光医学图像四组多谱段降质图像实验分析,以对比度和信息熵作为评价指标,与同类算法进行了图像增强效果的定性和定量对比,结果表明本文算法增强后的图像纹理和边缘细节更加丰富、对比度更高、视觉效果更佳,可广泛地应用于多种图像增强领域.According to irradiation-reflection model, by combining the generalized bounded operation model with guide filter, the problem of enhancement for multispectral degraded images with blurred details can be effectively solved and the contrast and low signal-to-noise ratio can be improved. The multi-scale reflection component image, i.e., final enhanced image is obtained through the following procedures： using the adaptive different scales of guide filter function as surround function estimate reaction; separating out the high-low-frequency information; obtaining the different scales Irradiation images which react the overall structure of the image; using the bounded generalized logarithmic ratio （GLR） model addition to replace the Retinex logarithmic transformation; taking a similar logarithmic transformation to the original image to improve the contrast of the image and make the dark area of image details enhanced; again using GLR model subtraction to remove illuminate components from the original image to segment the different scales of the reflection image, thereby avoiding the loss of small details and the big details caused halo effect and noise interference. With four direction Sobel gradient image which reflects the comprehensive edge details of image information the adaptive gain function can be obtained. To avoid the smooth area noise amplification~ by using the GLR model multiplication and addition to fuse the effective information of different scales images, the multi-scale reflection image, namely the final enhanced image are obtained. The effective suppression of the emergence of halo effect and computing overflow, which can retain a large number of image details; the comparision of subjective visual effect and the quantitative parameter analysis of the visible low illumination image, haze image, infrared image and X-ray medical images （a total of four groups of multispectral degraded images）, the use of the contrast and entropy as evaluation indices, qualitative and quantitative comparison wi

关 键 词：图像增强 照射_反射模型 有界运算 引导滤波 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]