Improved seam carving combining with 3D saliency for image retargeting

Seam carving is a content-aware image retargeting algorithm that removes the pixels with less energy values during resizing process to preserve important parts. Though many researchers have improved this algorithm by different ways, it is still a difficult problem to determine the energy function for removing task. Most existing approaches only use 2D features, and no single energy function performs well across all kinds of images so far. In this paper, we introduce a novel method that combine conventional L−1 norm of gradient with depth-aware saliency (3D saliency) to obtain energy map. Due to the different characteristics of these two operators in image analysis, our energy function contains both local and global information, which is proven to be effective in seam carving. The experimental results demonstrate the advantage of our method compared to conventional seam carving techniques.