基于多尺度Retinex的缺陷图像增强算法

摘要

An image enhancement based on multi-scale Retinex (MSR) is proposed to solve the problem of non-uniform illumination and poor contrast of defect image in detection of fluorescence magnetic particle. The principle and implementation of the algorithm are described. Firstly, the Gaussian filter coefficients are calculated for three different scales. Then, the original image is convolved and the results are weighted and averaged. Finally, the image is truncated and pulled up to get the result. The experimental results show that compared with the conventional algorithms, the MSR not only realizes the dynamic range compression, but also improves the contrast and improves the image quality.%针对荧光磁粉检测中缺陷图像光照不均匀、对比度差的问题,提出了一种基于多尺度Retinex(MSR)的图像增强算法,阐述了该算法的原理及实现方法.首先,计算三种不同尺度下的高斯滤波函数系数;然后分别对原始图像进行卷积操作并将结果加权平均;最后,将图像进行截断拉升,得到处理结果.试验结果表明:多尺度算法与常规算法相比,既实现了图像动态范围的压缩,又完成了对比度的提升,提高了图像的质量.