改进的高光边缘颜色恒常性算法研究

Research on improved specular edge color constancy algorithm

图像的高光区域包含丰富的光源信息,对颜色恒常性研究具有重要价值。然而,由于自然场景的动态范围往往会超过常用数码相机中传感器所能捕获的范围,高光区域易产生过曝光。针对过曝光现象,提出了一种过曝光区域恢复的高光边缘颜色恒常算法。该算法对高光边缘权重进行了2次优化。第1次优化在过曝光区域校正部分,利用过曝光区域自动校正算法恢复出高光区域的边缘信息,得到高动态范围图像;然后通过迭代算法将图像梯度投影到估计出的光源方向上,得到高光边缘权重的第2次优化;最后将优化后的高光边缘方向作为光源方向的估计。实验表明,对于自然场景图像,特别是其中过曝光明显的图像,该算法有效地解决了权重灰边颜色恒常算法中高光边缘不准确的问题,大大提高了光源估计的准确性。

The highlight regions of an image contain rich information of the light source, which contributes greatly to the study of color constancy. However, due to the dynamic range of natural scene often exceeds what the camera sensor can capture, the high- light regions are prone to over-exposure. Aiming at the over-exposure phenomenon, this paper presents an improved specular edge color constancy algorithm to recover the over-exposed regions. The algorithm optimizes the specular edge weights in two steps. The first optimization is in the over-exposure region correction part. The over-exposure region auto-correction algorithm is applied to recover the edge information of the highlight regions, and the high dynamic range image is obtained. Then, with the iterative algorithm, the gradient of the image is projected onto the estimated direction of the light source, the second optimization of the specular edge weights is achieved. Finally, the optimized specular edge direction is taken as the estimation of the direction of the light source. Experiment results show that for the natural scene images, especially for the obviously over-exposed images, the proposed algorithm effectively solves the inaccurate specular edge problem in the weighted grey-edge color constancy algorithm, and greatly improves the accuracy of the light source estimation.