基于最佳分析窗口的高光检测

物体上的高光直接影响计算机视觉领域的很多操作,比如图像分割、边缘检测、视频中物体的跟踪等.该方法基于双色反射模型,提出了最佳分析窗口的概念.在最佳分析窗口的基础上,分析窗口中像素的亮度、色彩、以及位置信息,组成五维特征向量,利用K-means算法在向量空间的聚类结果得到漫反射像素族,从而估计出全局漫反射色度,并以此检测图像中的高光像素.最后,利用Tan等提出的STD机制从原图像中分离出漫反射分量和镜面反射分量.该方法无需交互操作,不受相机噪声为线性的限制,比同类算法更容易实现且效果更佳.

Highlight detection based on best fit window

The existence of highlight reflected from inhomogeneous objects always produces errors in computer vision's algorithms,e.g.image segmentation,edge detection,object tracking in video.This method bases on DRM and Best Fit Window is proposed.In Best Fit Window,diffuse pixels cluster is obtained by using K-means algorithm in the five dimensions feature vector space: CIE L*a*b in-formation and pixel location information.And then the highlight cluster is detected.Finally,diffuse and specular reflection components are separated by using STD mechanism which is proposed by Tan et al.This algorithm is easy to implement,free from the limitation of linear camera noise and requires no interactive operation,but efficient.Experiments show that it also partly works when highlight exists on several different objects.