An Image Object Segmentation Algorithm Based On Particle Filters

Image object segmentation, to segment the specific object from clustered background, is the ultimate goal of image segmentation. Since image segmentation technique has been widely used in almost every field related to image, image object segmentation is a worthy field to investigate.According to human vision, image object segmentation requires the prior knowledge of object's feature as well as the low-level information from image. Thus, increasingly number of researchers start to combine the bottom-up and top-down processes to solve the problem. In this paper, a similar method is applied. Low-level information is obtained from the Normalized Cuts, a typical image segmentation algorithm; while a partial shape dataset is built to provide high-level information. With the model of particle filters, the two level information is combined successfully. The experimental results demonstrate its validity.The main contents of this paper are as follows1)Taking Normalized Cuts for example, we discuss the definition and category of image segmentation algorithm. We use edge fragments from the Normalized Cuts to gain low-level information.2)Due to the variation of the object's shape, we use the partial shape feature to provide high-level information, instead of the traditional contour/region based shape descriptors.3) The algorithm uses particle filters model to combine the two level information. Each edge fragment in Normalized Cuts image is modeled as a particle, whose weight updates as it grows longer. The algorithm converges to the object's contour.The experimental results verify the capability of particle filters model for successfully combining the low-level information from the image and high-level partial shape characteristic from the dataset. Another advantage of this algorithm is its excellent performance against large texture variation and partial object occlusion.