The Research Of Image Resizing Based On The Content Of The Image

Image resizing is one of the most important parts of image processing, conventional methods use uniformly scaling images to fit the new target screen. At present, most cellphone and PDAs adopt this solution, but the images maybe shrink so much that important parts can not be recognized or distorted during the resizing procedure; Cropping approaches only emphasize the important parts and just discards the surrounding to fit the display medium, this almost discard large number of background information, which lead to user's misunderstanding of the image. Image resizing based on content is now one of the research focuses, which considers the image content in the process of image resizing.In chapter1, we introduce approaches to uniformly scaling and cropping of images. As well, we emphasize some algorithm particularly on content-based image resizing, such as Fisheye-View Warping, Seam Carving, Segment and pasting, Random walk model for content-Aware video resizing, optimized scale-and-stretch and triangle grid parameter method and so on, we also analyze the merit and demerit of these algorithms.Image resizing based on content involves calculation of saliency, for each pixel or image block, a value of how much it attracts the visual system. In chapter 2, we introduce the calculation of image salience map and some basic knowledge of digital image processing. The finite element is used to resize a image in the thesis, so the finite element and mesh generation are also mentioned in this part.To keep the image integrity and protect the important edge, and make the viewers grasp better the important information, we propose an approach based on finite element method to image resizing, wherein the characteristic tri-mesh is exploited for image division. In chapter 3, we focus on how to generate the mesh and how to set the stiffness matrix. Different stiffness is set to different element according to the saliencies of triangular edges, so that when given the same power, different edge will stretch or shrink differently. Thus resize the images non-linearly according to the saliency of the areas.In chapter 4, some examples are illustrated applying the proposed algorithm on different kinds of images, which concludes the thesis with discussions of the digital results and limitations of the presented method.