Foreground Object Extaction Based on Independent Component Analysis

Moving objects are often characterized by coherent motion that is distinct from that of the background. This makes motion a very useful feature for segmenting video sequence. Extracting moving objects from videos is important for many applications like surveillance, traffic analysis etc. In this paper a novel and efficient moving object segmentation algorithm is proposed that is based on independent component analysis (ICA). Moving objects and static background are considered to be independent, so independent component analysis is applied on frames of a video sequence to identify the preliminary independent components containing moving objects. This source image data obtained after ICA are further processed using anisotropic diffusion. Anisotropic diffusion is used here to reduce the noise present in the preliminary source image without removing the significant parts of the image content. Finally, a post-processing step based on morphology is applied on the obtained objects to remove small unnecessary objects and to smooth the object boundary to produce the final segmented images indicating the moving objects. The method is tested on various datasets and experimental results establish the satisfactory performance of the proposed algorithm.