Optimal bi-directional seam carving for content-aware image resizing

Abstract

An optimal bi-directional seam carving method for content-aware image resizing is proposed in this paper. It can adaptively confirm the quantity of the carving seams in both horizontal and vertical directions according to the homogeneous region distribution obtained from the significance map. The significance map is a weighted average of the gradient magnitude and saliency measure which is calculated by our multi-resolution saliency model. Comparing the previous unidirectional seam carving methods of backward and forward, our method fully utilizes the available homogeneous information in arbitrary position of the original image. Therefore, it will avoid to delete or insert seams passing through the prominent features and structure objects so as to decrease or even eliminate the artifacts of staircase effects. © 2010 Springer-Verlag Berlin Heidelberg.