Research Co-segmentation On Images

Image segmentation is an important research topic for image processing and computer vision. It plays an important role in many applications, such as motion capture, image retrieval, object tracking,3D reconstruction, video analysis and video compression, etc. Co-segmentation is a method that makes use of the similarity among multiple images and simultaneously segments common objects from multiple images. Compared with traditional image segmentation, co-segmentation can more accurately simultaneously segment common objects from multiple images, and requires less human intervention, thus it has attracted the attention of many scholars, and has achieved remarkable results. However, due to the complexity and diversity of images, the robustness of co-segmentation algorithm need to be further enhanced.This paper studies the commonly used energy functions and optimization models, and proposes a co-segmentation algorithm which is based on regional active contour model, and it can simultaneously segment three images. Compared with the traditional active contour model, our model added the constraint of the foreground similarity among images and the constraint of background consistency in each image. By analyzing the commonly used models for foreground similarity measure, the rewarding strategy is more convenient for optimization than punishment strategy. In order to measure the foreground similarity better, we update the traditional rewarding strategy. We replace the region representation of Chan-Vese model, and select the color histogram as the region representation instead of mean value. According to the proposed energy function to generate the level set, we use Euler-Lagrange equation to minimize the energy function. It makes the contour curve converge to the common object boundary.Apart from the segmentation of images with the similar foreground region, co-segmentation can also be used for segmentation of multi-view images. The multi-view co-segmentation can more accurately segment object, moreover the results of segmentation can be used for3D reconstruction. Compared with some previous multi-view image segmentation algorithms, co-segmentation is without camera calibration and just need less user intervention. It’s more convenient to operate.In this paper, we use the mixed programming of C#and matlab to build a platform of co-segmentation. We segment the commonly used image database and project database, and get a better result.