Gradual Color Clustering Elimination as a Novel and Efficient Method for Outdoor Image Segmentation

摘要

Automatic processing of outdoor images as one of the most important issues in computer vision system has variety of applications in traffic control, robots navigation, and assisting visually impaired people for safe travel. The low-level features analysis, presents an enormous challenge to the segmentation of the outdoor images since the special characteristics of such images are subject to changes (for example luminance effects and color/texture variety). Moreover, large and small objects often result in either over-segmentation or under-segmentation. This study is an attempt to customize the color clustering methods for segmentation and object recognition in the outdoor images, using a multi-phase procedure through a multi-resolution platform, called Color Cluster Elimination (GCCE). In each phase, the primary color clusters are detected and then gradually eliminated, to allow the smaller clusters to emerge in much clearer versions of the image. The proposed method can not only overcome the segmentation problems but also build up substantial resistance against noise. The major clusters are identified using the adaptive statistical method and color histogram analysis. Moreover, the morphology functions are used to eliminate the sub-regions and the weighted graphs are employed to estimate the number of the primary colors and merge the adjacent homogenous regions. The proposed method has been evaluated on outdoor images dataset namely BSDS and the results have been compared to PRI and GCE statistical metrics of the latest segmentation methods. The comparative tables show the proposed method has got premising performance for the segmentation of outdoor image.