摘要
指出了Retinex彩色图像增强理论的物理意义,在对数空间中,将原图像减去高斯函数与原图像的卷积,其物理本质是除去了原图像中的平滑的部分,突出了原图像中的快速变化的部分,而且高斯函数越尖锐,越是突出图像中的细节,高斯函数越平坦,图像色调保持得越好,多尺度Retinex综合了不同高斯函数与原图像进行卷积的优点。研究了多尺度Retinex标准差截断法,结果是,以多尺度Retinex处理后的图像强度在其平均值附近1倍标准差截断后再拉伸得到的图像,普遍好于以2倍、3倍标准差截断后再拉伸得到的图像。
The physical idea of Retinex theory that used in color image enhancement was revealed. The product of the convolution between a Gauss function and the original image, which was the smooth part of the original image, subtracted the original image in logarithm space. The left part was quick transform part of the original image. The details of the original image were highlighted. The sharper of the Gauss function was, the more details were highlighted. The smoother of the Gauss function was, the better hue of the image was showed. The result of multi-scale Retinex （MSR） has advantages in different singlescale Retinex （SSR）. The method of standard deviation in intercepted region of handled image by MSR was researched. The result is obvious that the image quality intercepted in the region of [μ-σ ,μ-σ ], which μ is mean of image handled by MSR, is better than intercepted in the region of [μ - 2σ,μ - 2σ] and [μ - 3σ,μ ＋ 3σ].
出处
《光学技术》
EI
CAS
CSCD
北大核心
2007年第1期127-129,共3页
Optical Technique
