用于去除单张图像高光的光照约束补色

在计算机视觉研究领域,如何检测和消除图像中的高光(specular)一直是个热点问题,有关的研究结果对于提高计算机视觉算法性能有着重要的影响.针对这一问题,提出了一种检测和消除高光的方法.首先,通过比较高光和漫反射光(diffuse)的色度特性的不同,给出了一种交互检测单色物体表面高光区域的方法;然后,引入补色(inpainting)方法并结合光照约束条件,设计了一种去除单张图像中高光并还原出漫反射分量的新的补色算法.与一般补色方法不同,该算法充分利用了高光区域含有的信息来指导补色过程.通过综合利用观测到的像素值、光源的色度分析(illumination chromaticity analysis)、光源颜色的平滑性等来约束补色过程,保证了算法能够克服一般的补色方法无法保持物体表面细微明暗变化的缺点.实验结果表明,与以往的去除单张图像高光的方法相比,该算法能够提供更好的光源色度估计,从而得到更准确的结果.

Illumination-Constrained Inpainting for Single Image Highlight Removal

Specular detection and removal are always hot problems in computer vision. Advanced results have a great impact on computer vision algorithms. In this paper, a specular detection and removal algorithm is proposed. First, through the comparison between highlight and diffuse chromaticity, a user interactive detection method for a single colored surface is developed. Second, a removal algorithm is proposed by introducing an image inpainting algorithm to this field and adding an illumination constraint. Different from traditional inpainting algorithms, the method described in this paper utilizes more cues which are embedded in the specular region. By integrating the embedded information such as the observed pixel color and illumination chromaticity, this method can overcome the shortcomings of the traditional inpainting methods, which can not keep shading variance in the highlight region. Experimental results show that this method can give a better illumination chromaticity estimation and more convincing results.