Fast Content Aware Image Retargeting

This paper addresses the problem of retargeting, namely adapting large source images for effective viewing at a smaller size with possible applications to PDAs, or dynamic page layouts. Instead of extracting regions of interest for retargeting, the uninteresting parts are removed from the scene [8]. This is done by computing the RGB variance within non-overlapping 3x3 blocks and removing the block path with minimal variance cost using dynamic programming. It is shown that transformation to CIELAB space is more effective for visual interpretation of image content. The implementations are shown to be much faster than the seam carving approach of [8]. Schemes are also presented for speeding up the seam carving scheme itself.