| In the computer vision theory, the image segmentation result is directly related to the feature extraction and object recognition. The separation of the object in the image and other background information, has always been the basic problem in computer vision. But the traditional image segmentation is to partition the image into similar regions, it will be difficult to separate the objects from the background region. So there are defects of the subsequent calculation relied on the object in the image. An image segmentation method based on visual attention, is proposed to approach the defects described above.The main work and innovations are as follows:1.We propose a new visual attention computational model based on the sparse and color information. Not only it can improve the effects of the noise on the sparse transformation, also can overcome the excessive dependence of color information. In the MSRA image sets, compared with solely on frequency sparse algorithm or color feature algorithm or other algorithms, our results show that the method has good results.2.Combined with the new attention algorithm, we propose an image segmentation algorithm based on automatically region merging. Firstly, we use the Mean Shift algorithm to obtain more similar regions. Then we mark the target regions using the prior obtained by space saliency information. Finally, we merge with the maximum similarity of target areas and get the final significant objects. Compared with the Ground_Truth and the traditional segmentation algorithms, the experimental results on MSRA image sets have confirmed the validity of the new algorithm.3.In order to compare and analyze the experimental results, we designed and developed a set of evaluation system. The software shows the saliency results of the proposed algorithm and other algorithms, and can also analyze and evaluate of the results by different evaluation criteria. In addition, it displays the results of image segmentation and the similarity of the results. |
PDF Full Text Request