Abstract: Medical X ray images are usually difficult to identify because of the poor brightness and low contrast. In order to enhance these images, this paper discusses the Multi-Scale Retinex（MSR） image enhancement method on the basis of the Single-Scale Retinex put forward by Land. A fact is emphasized that the calculation in image filtering can use a mean template as an alternative to Gaussian convolution template, and an improved method of canonical gain/offset is proposed to map the image gray-level on display devices in this paper. Further experiments compare the new method based on Multi-Scale Retinex with histogram equalization and gamma correction. It can be seen from the experimental results that the proposed method enhances image brightness and contrast ratio compared to the other two mentioned methods, and the entropy of X ray images is improved by the new introduced method; therefore the algorithm can satisfy the demand of medical image diagnosis.

Key words: X ray image, Multi-Scale Retinex（MSR）, Gaussian convolution, mean template, map the image gray-level

摘要： 医学X射线图像通常存在亮度低，对比度差，造成难以识别的问题。为了增强此类图像，在Land提出的单尺度Retinex理论之上，论述基于此原理的多尺度Retinex（MSR）图像增强方法。利用均值模板代替高斯卷积模板对图像进行滤波，并且改进了将图像映射到设备显示器上的canonical gain/offset修正方法。实验将改进的修正方法用于多尺度的Retinex之上并与直方图均衡化和伽马校正方法进行对比。实验结果表明图像亮度的增强和对比度提高优于上述其他两种方法，新提出的方法较原有方法有效地提高了图像的信息熵，满足医学图像的诊断需求。

关键词: X射线图像, 多尺度Retinex, 高斯卷积, 均值模板, 图像灰度级映射