Image retargeting based on delaunay triangle mesh

Accurate presentation of high resolution images on wide varieties of multimedia devices, especially on pervasive mobile multimedia devices, faces great challenge. In this paper, we present an angle preserved warping method for image retargeting which allows resizing images to arbitrary aspect ratios while minimizing the loss of important information. To accomplish this, we first cover the original image with a Delaunay triangle mesh based on saliency map of the image, which is computed by considering the lumination, color, gradient and position features of the pixels comprehensively; and then the triangle mesh is deformed to target aspect ratios subject to a deformation loss function, to drive the visual prominent features to undergo similarity transformations. The deformation loss function is constrained by the angle's changes of the triangles in the triangle mesh. Compared with previous approaches using grid mesh, our method can transfer the distortion caused by the change of the image aspect ratios to less saliency regions much more effectively.