基于混合能量的内容敏感图像缩放新方法

Content-Aware Image Resizing Based on Hybrid Energy

提出了一种内容敏感图像缩放的新方法.该方法旨在保持图像缩放前后重要特征区域尽量相似的前提下,通过约束重要区域的旋转使得图像缩放前后视觉效果尽量一致.该方法分为3个步骤:首先,为图像关联全局和局部的控制单元,并根据这些控制单元所覆盖原始区域的重要性给其赋予权值;然后定义一种新的混合能量来约束控制单元,在变化前后尽量保持相似并惩罚其旋转;最后通过优化混合能量加权和来计算这些控制单元变形后的最佳位置,并插值得到最终图像变形结果.大量的对比实验结果表明,该方法具有保全局和局部特征、重要物体不易旋转、能够高效地得到扭曲较小的结果等优点.

This paper presents a new content aware image resizing method. This method tries to keep important features as similar as possible to their origin, and aims to provide similar visual appearance after resizing by reducing rotation of important areas. This method consists of three steps： Associate global or local control units for the image, and assign weights according to the regions they cover; define new hybrid energy to constrain the units, making them as similar as possible before and after deformation with rotation penalized. These energy items are finally add- ed and minimized to get the positions of the deformed control units, and the new image is ob- tained by interpolation. A lot of comparison results show that the method can keep both global and local features, avoid over rotating important objects, and the final results are less distorted and better than previous methods.