Multi-operator retargeting for stereoscopic images towards salient feature classification

Abstract

Most stereoscopic image retargeting (SIR) algorithms often use a single operator or a fixed strategy to resize various images. They ignore the adaptation between retargeting methods and specific image features, hence they fail to achieve a desirable retargeting quality. We propose a multi-operator retargeting method for stereoscopic images based on salient feature classification to address the issue. Specially, the original stereo image is first classified according to its salient features, including spatial and depth salient features. Then three operators, including stereo cropping, stereo seam carving, and stereo uniform scaling are combined to perform image retargeting in terms of image category, salient features, and target size. In particular, we design a retargeting strategy used to realize adaptive switching between operators. Besides, we construct two distance energy terms and integrate them into the total energy function of stereo cropping and stereo seam carving respectively to improve the quality of retargeted images. Extensive experiment results show that the performance of the proposed method is superior to that of other SIR algorithms. The proposed method can effectively preserve the integrity and geometry of the salient content while ensuring the stereoscopic sense of the images during retargeting.