Automatic moving object segmentation using histogram-based graph cut and label maps
Daum Kim, Joonki Paik · Electronics Letters · 2012
Presented is a novel object segmentation method using a histogram-based graph cut algorithm with automatically generated label maps. The proposed method consists of three steps including: (i) pre-processing, (ii) label map generation and update, and (iii) object segmentation. The pre-processing step over-segments the input image using the median filter-based watershed algorithm. The label map is generated by using optical-flow and morphological skeletonisation, and it is updated using colour histogram and the Bhattacharyya coefficient for the following frame. Moving objects are finally extracted by using the histogram-based graph-cut method. Experimental results show that the proposed method is significantly faster than existing object segmentation methods by removing user interaction.