自适应补偿Retinex算法的多尺度图像渐晕校正

Multi Scale Image Vignetting Correction Based on Adaptive Compensation Retinex Algorithm

图像渐晕会干扰机器视觉处理的性能,考虑到现有图像渐晕校正出现的失真、泛灰等质量问题,设计了自适应补偿Retinex方法。根据图像渐晕模型分析,图像渐晕主要由于径向像素强度变化引起,而这正好符合Retinex理论的分解原理,于是引入Retinex算法削弱亮度分量,恢复渐晕图像的入射分量。为避免图像失真,设计了多尺度Retinex,同时考虑到尺度取值会影响亮度与对比度、色彩之间的均衡,引入控制因子抵消尺度影响。为了消除光晕和失真,采用区域生长和灰度调整机制,自适应补偿渐晕边界。仿真从主观验证了自适应补偿Retinex方法适用于不同类型的渐晕图像校正;从客观图像角度验证了方法具有更好的校正图像质量和细节纹理;从光谱角度验证了方法校正后的图像光谱更为均衡,与原始图像具有更好的一致性。

Image vignetting will interfere with the performance of machine vision processing. Considering the quality problems of existing image vignetting correction, such as distortion and graying, an adaptive compensation Retinex method was designed. According to the analysis of image vignetting model, the image vignetting is mainly caused by the change of radial pixel intensity, which is in line with the decomposition principle of Retinex theory. So Retinex algorithm was introduced to weaken the luminance component and recover the incident component of vignetting image. In order to avoid image distortion, a multi-scale Retinex was designed. Considering that the scale value can affect the balance between brightness, contrast and color, a control factor was introduced to counteract the scale effect. In order to eliminate the halo and distortion, the region growing and gray adjusting mechanisms were adopted to compensate the vignetting boundary adaptively. Simulation results show that the Retinex method is suitable for different types of vignetting image correction;from the perspective of objective image, it is verified that the method has better correction image quality and detail texture;from the perspective of spectrum, it is verified that the spectrum of the corrected image is more balanced and has better consistency with the original image.