尺度变化的Retinex红外图像增强

经典Retinex模型增强算法采用固定尺度高斯核平滑滤波,导致单一尺度Retinex无法进行全局有效增强,而多尺度Retinex权重系数选取困难,二者均不能满足视觉要求。针对以上问题,基于人眼视觉掩盖效应提出一种尺度变化高斯核平滑滤波的Retinex算法。首先利用人眼视觉掩盖效应的屏蔽函数检测像素邻域空间细节,依据像素区域细节信息丰富程度设计出尺度变化的高斯平滑滤波器,实现照度估计,最后对尺度变化高斯平滑滤波器实现提出实用方法。实验证明本文算法有效提高红外图像对比度,增强细节信息,在主观视觉效果和客观评价指标上整体优于修正对比度限制直方图均衡算法、单尺度Retinex、多尺度Retinex及平稳小波和Retinex增强算法。

Infrared image enhancement method based on scale varies Retinex theory

Classic Retinex algorithm, applied in image enhancement with fixed Gaussian filter,couldn't meet demand of human vision. The result of Single Scale Retinex algorithm was unable to satisfy effective enhancement of the whole image, and it was difficult to use Multi-scale Retinex algorithm estimating illumination correctly without suitable coefficient. As a result, a novel scale varies algorithm of Retinex theory based on human visual masking effects was proposed. The results of human visual masking function, which can estimate how much details exist in the pixel neighborhood efficiently, was used to design the scale of Gaussian filter. The new algorithm was designed to get the correct illumination of a infrared image. At last, a practical solution was presented to cut down time consuming.Experimental results show that the proposed algorithm improves the contrast and enhance the details of infrared images. Moreover, the proposed algorithm is better than histogram equalization, Modified Contrast Limited Adaptive Histogram Equalization, single scale Retinex, Multi-scale Retinex and Stationary Wavelet and Retinex in subjective visual effects and objective evaluation parameters.