Comparison of Moving Object Segmentation Techniques

Abstract

Moving object segmentation is the extraction of meaningful features from series of images. In this paper, different types of moving object segmentation techniques such as Principal Component Analysis (PCA), K-Means clustering (KM), Genetic Algorithm (GA) and Genetic Algorithm Initialized K-means clustering (GAIK) have been compared. From our analysis we have observed that PCA reduces dimension or size of data for further processing, which in return reduces the computational time. However, the segmentation quality sometimes becomes unacceptable. On the other hand, due to random initialization of its centroids, KM clustering sometimes converges to local minimum which results in bad segmentation. Another algorithm which has been considered in this study is GA, which searches all the feature space and results in a global optimum clustering. Although the segmentation quality is good, it is computationally expensive. To mitigate these problems, KM and GA are merged to form GAIK, where GA helps to initialize the centroids of the clustering. From our study, it has been found out that GAIK is superior to GA in both the quality of segmentation and computational time. Therefore, in general, the analyses of the four algorithms shows that GAIK is optimal for segmenting a moving object.