(a) The input image (the fish is the salient object); (b-f) Resized images with different scales of the salient object.
Abstract
We present a novel
approach for interactive content-aware image resizing. The resizing is
performed on warping a triangular mesh over the image, which captures
the image saliency information as well as the underlying image features.
The warped triangular mesh and the horizontal and vertical scales of all
triangles are simultaneously obtained by a quadratic optimization which
can be achieved by solving a sparse linear system. Our approach can
preserve the shapes of curved features in the resized images. The
resizing operation can be performed in an interactive rate which makes
the proposed approach practically useful for realtime image resizing. To
guarantee foldover free resizing result, we modify the optimization to a
standard quadratic programming. A number of experimental results have
shown that our approach has obtained pleasing results and outperforms
the previous approaches.
Resizing results with preserving features like lines (upper row), curve (middle row), and circle (lower row). From left column to right column: the original images, the resizing results without preserving features, the specified features with sampling points shown in orange, the resizing results with preserving features..
Thanks to the many
Flickr.com users whose images were used in this paper. This work is
supported by the National Natural Science Foundation of
China(60776799,10871178), Technology Department of Zhejiang Province
(No. 2008C010483), and the 973 National Key Basic Research Foundation of
China (No. 2009CB320801).
title = {Nonhomogeneous Scaling Optimization for Realtime Image Resizing},
author = {Yong Jin and Ligang Liu and Qingbiao Wu}
journal = {The Visual Computer (Proceedings of CGI)},
volume = {26},
number = {6-8},
pages = {769-778},
year = {2010}
}