Research On Content-aware Image Retargeting

In recent years, the display devices variincrease, it requires that an image or a video should be displayed on the display devices of different size or aspect ratio. The traditional scaling or cropping has not considered the content of the image or video, the results are not satisfying. Recently, content aware retargeting is becoming a hot point. In this paper, our work focus on the image retargeting. Two approaches are proposed. The details are included as follows:1、Considering the complementary features of the continuous and discrete approaches, we proposed a hybrid approach to combining seam context constrained seam carving and scaling.First, addressing the problem of artifacts and time consuming of seam carving, we propose seam context constrained fast seam carving, so that the removed seams distribute more scattered. It causes fewer artifacts in the resized images. Furthermore, we proposed Content Aware Measure (CAM) to measure the similarity between the original and the retargeted images. If the CAM is over a threshold, seam carving is switched to scaling. The compared experimental results show that the proposed approach has discriminate improvements in both image quality and running time.2、We propose a non-homogenuous mapping approach to optimize mesh defomation. First, the image is segmented into several meshes by the image importance. Then, the meshes are deformed into the target size by the importance. Finally, the target image is obtained by warping. The proposed non-homogenuous warpping framework is simple and effective. It can preserve the important regions with fewer artifacts. Furthermore, a novel optimal model is proposed to initialize the grid mesh partition by the importance. It reduces the deformation of the contour of the key object and improves the image quality.