Research On Content-Aware Image Resizing Based On Seam Carving

Image resizing is a basic digital image processing technology. Especially in today's rapid development of information technology, there are a lot of different resolution and aspect ratio electronic products. When passing images among them, how to retarget image to meet these display medium that maintaining the salient objects intact. Therefore, human require for new and higher quality service. They expect to have better experience of digital entertainment. In recent years, image resizing technology for different application becomes a hot topic. They all try to maintain the visual prominent objects intact during resizing process, which is called as the content-aware image resizing.In this thesis, we proposed a new seam carving method based on salient map and gradient magnitude. Then a multi-operations method is completed, which are based on seam carving resizing. Finally we complete an interaction image retargeting based on multi-operations. Users can choose her or his favorite resulting image. The main contributions are:(1) To maintain the visual prominent object intact, the energy cost function combine of gradient magnitude and salient map, can prevent seams passing the visual prominent object.(2) A better resizing result is gained by forward energy standard, which considering not only the energy itself, but also the energy which induced into the images by applying the operation.(3) Through the user studying, they can't accept the distortion caused by seam carving, while preferring crop operation to scale operation. So we should prevent seam carving operation in time even in advance when it damage the prominent object during the multi-operations image resizing process. (4) Dominant color descriptor (DCD) is adopted to measure damage of each operation to the global visual effect. We adjust the energy cost function with a new parameter, in order to prevent the seam carving operation from damaging the prominent object in time or in advance. In the interactive method, user can choose her or his favorite one from many candidate results.