Object Segmentation

  • Object segmentation

Object segmentation is one fundamental process in computer vision and frequently used in image-processing applications to make images easier to analyze by labeling every pixel whether object or background. Traditional object segmentation algorithms require additional information which usually specified by users, i.e. manually drawing strokes on the object and background regions separately or grabbing a rectangle box around the object, to modeling initial color model of object and background. However, the user interaction is time-consuming and should be performed on every image respectively. So, it is hard to deal with massive dataset or real-time process. One of substitute methods for manual input is using a visual saliency map which can be obtained automatically. We propose a novel approach of building initial model based on saliency map. By using saliency map without binarizing process, the proposed algorithm shows better performance even without additional information.