A novel SVM video object extraction technology

Abstract

For the problems of fuzzy object's edges and computation complexity for video object segmentation, an improved SVM algorithm is proposed in this paper. We have adopted the adaptive change detection method to get the original video object, whose pixels constitute the samples set for SVM training, and then we improved the SVM by using the idea of active learning, and finally we built the video object segmentation model from the improved SVM. Experimental results show that both the spatial accuracy and the temporal coherency of this algorithm are much better than before. This algorithm achieves the goal of automatic segmentation, and overcomes the disadvantage of supervision learning, and it can reduce the computation complexity.