Research And Application Of Non-Uniform Warping For Content-Aware Image Retargeting

With the proliferation of display devices, adjusting an image to heterogeneous devices with different sizes and aspect ratios is becoming more attractive. Image retargeting is the process of adapting an existing image to display with arbitrary sizes and aspect ratios, for which the critical problem is how to retarget the image effectively and to prevent the prominent object of the image from distorting. The most common retargeting method is resizing images to some give dimensions, the most simple retargeting algorithm is uniform scaling, but in most cases it will stretch and distort the significant content and the significant features will become too small to be recognizable on small screens. To address this problem, researchers have spent a large amount of effort on content-aware image retargeting method. Content-aware is a way to protect significant content and features during the retargeting process, and avoid visual distortion and stretch at the same time.According to the analysis that the important map and the mesh partition fine degree of an image will affect the target image, in this paper, we present a novel image retargeting method based on non-uniform mesh warping for content-aware, of which main idea is first sampling mesh vertices based on the saliency map, then defining quadratic error metrics to measure image distortion in order to preserve the important content and the global visual effect of the image, and finally formulating the image retargeting as a quadratic minimization problem carried out by solving linear systems.In this paper, we adopt a new method to compute the important map which will detect the mesh partition, and its process includes two steps:(1). compute the saliency map, face information and edge information by using GBVS method, face-detection and Canny operator respectively;(2) get the final important map by combining these information. This method can not only preserve the important region, but also make up the disadvantage caused by GBVS method.Due to the mesh partition fine degree have an important effect on the target image and the computing efficiency, in this paper, we propose a new mesh partition method based the important map, and its core idea is to embed the important region in the fine meshes and embed the unimportant region in the coarse meshes. Moreover, we introduce an interactive way to assign the important region because which part is important or unimportant is subjectivity and the important region which is computed automatically may is not necessarily that users think.For the problem that the target image will suffer from serious distortion for the sake of the global visual effect is easy damaged during the retargeting process, we use a non-homogeneous warping for image retargeting method. In this method, we apply mesh-linking scheme among similar neighborhood meshes in order to better preserve the global image structure. Besides, we use different strategies for upsizing and downsizing, which can get the target image accord with the target display. Final we get the target image by adopting the texture method and compare our results with other methods’. And, our experimental results verify that our method can better preserve the important feature and global visual effect in the image.