Study On Content-Aware Image Partition And Resizing

The diversity of display devices and the development of digital network require that image and video could be fairily displayed with different resolutions or aspect ratios in various display devices, simultaneously the region of interest and important information should be protected effectively. In this paper, based on image gradient information, image strip dividing and feature information of various objects in the image, content-aware image resizing method is researched.A fast image retargeting algorithm based on strips dividing is studied. Firstly, the image energy map is constructed by gradient information of pixels, for describing the importance of the regions and objects in the image. The image can be partitioned into several strips by integrating the similar energy-level columns together. The reduced amount of each strip is determined according to its accumulated average energy and the resizing amount of the image, and the image strip is scaled in terms of its reduced ratio. The strip is cropped when the reduced ratio exceeds a setting threshold. Experimental results show that the algorithm is capable of preserving both the local structures and the global visual effect of the image, producing the high-fidelity retargeted results.Based on the feature detection technology of image salient object, an image retargeting algorithm is researched. A cascade of boosted human face classifier, which is used to detect the human face feature of an image, is trained by using7000images as positive samples and5000images as negative samples. The human body feature is also detected using the histogram of oriented gradient feature of image. The region of human feature in the energy map is weighted. Using strip dividing algorithm to realize image retargeting. An system of image resizing has been exploited to realize content-aware image resizing.On the basis of analyzing temporal coherence and estimating motion vector of the video frame, an efficient video retargeting algorithm is studied in this paper. First the energy map of the frame is constructed combining the gradient information of pixels with the motion vector calculated by block matching. Then in terms of the energy map, a single frame is partitioned into several strips, which the width of strip in current frame is determined by motion vector of pixels and the partitioned result of previous frame. The reduced amount of each strip is determined in inverse proportion to its average energy, and the corresponding strips across two adjacent frames ought to be scaled with as consistent ratio as possible to avoid the frame-to-frame jitter. After scaling the strips of the frame, we can get a resized frame, such that the retargeted video could be obtained by synthesizing all the resized frames. Experiments show that the proposed algorithm can be used to preserve the salient object and temporal coherence among frames, resulting in a fluent and jitter-free retargeted video.