Automatic Detection Specular Reflection Components in Leaf Images based on GMM

摘要

Specular reflection components can cause erroneous results on segmentation and recognition in vision algorithms. Forest inventory produce a lot of leaf images in natural light. Some of them have specular reflection components which cause a major obstacle in the way of automatic segmentation of such images. This paper proposes a method based on Gaussian Mixture Model (GMM) to detect the specular reflection components in leaf images. The method can separate the highlights from its neighborhood accurately. First, the coarse specular reflection components are found using Retinex Algorithm. Second, using RGB color channels of the coarse specular reflections individually to train the GMM and get the final highlights. The experiment results show that the proposed method can find almost the specular reflection components.