The Research On Content Perception Based Image And Video Resizing Techniques

When viewing digital image and video on mobile terminals such as mobile phones, PDA and MP4, it is necessary to make adjustment to the original image and video so as to be compatible with the resolution of display terminals. However, classical resizing techniques are far from satisfactory. For example, the image content will be obscue or be distorted. Though cropping can retain most important information, it will lose too much secondary information.In recent years, content perception based image and video resizing techniques are becoming the mainstream approaches in the field of video adaptation. After a brief overview of the recently available content based resizing techniques, two improved methods are proposed for the retargeting of digital image and video.First, after analyzing the most famous seam carving technique, an improved image resizing approach, which is based on saliency map and adaptive energy line, is proposed. It aims at overcome the limitation of energy function definition and the complexity of dynamaic programming. It makes the energy description of image content perfect by incorporating the saliency map based on visual salience model. Meanwhile, a content-based adptive coarse and fine energy line, together with hirechical acceleration based dynamic programming (HADP), is proposed to improve the efficiency of searm carving. Experimental results show that the proposed approach can not only effectively protect the important image regions, but also reduce the time consumption.Second, an image and video sampling technique is proposed based on importance map. The importance map is built from three basic properties of image and video, which can reflect the importance of every pixel in an image and video frame. And then, the importance diffusion function and adaptive sampling function are introduced. The importance diffusion function spreads the importance of those removed pixels to neighboring pixels so as to avoid deleting much too unimportant regions, which will results in the distortion of image's whole content. The adaptive sampling function defines the weighting coefficients in terms of the importance of pixels in every line and column, and thus protects the important image regions. By combining the advantages of these two fuctions, an energy function for equilbrim scaling is defnied. It trades off between the global content and important regions of digital image. Furthermore, the proposed approach can be utilized for video retargeting when taking the temporal coherency into onsideration. Experimental results demonstrate that the proposed approach is superior to those existed in the literature. It can not only effectively overcome the limitation of ambiguity in important regions by sampling, but also keeps the global content of image and video.