Content-Aware Retargeting of Stereoscopic Images

초록

In this paper, we propose a novel warping-based method for content-aware image resizing of stereoscopic image. The previous algorithm for single image retargeting does not consider the consistency between stereoscopic image pair. As a result, the geometric structure of independently retargeted images becomes distorted. On the other hand, the proposed algorithm retains the stereo consistency by matching vertex of grids in stereo image pair. Each vertex of grids is considered as a feature point in left image of the stereo pair, and the corresponding point is found on right image. By applying the same warping on the corresponding points of the image pair, stereo consistency is preserved. In addition, in order to improve the probability of preserving important object, we employ the sparse disparity map. On the vertices of grid, disparity is computed using the distance between corresponding feature points. Then, we utilize the GPU interpolation to generate the sparse disparity map. Experiment result shows that the proposed method retains the geometric structure, and salient objects are preserved properly in the stereoscopic image pairs.