Distortion-Aware Image Retargeting Based on Continuous Seam Carving Model

Abstract The seam-carving algorithm is a classic context-aware image retargeting method and has been extensively studied for years; however, distortions exist as the discrete iteration of least- energy computation. We propose a continuous seam carving model through the just noticeable distortion (JND) detection at every iteration and accumulative energy weight. The JND of every pixel is calculated by the minimal just-noticeable distortion energy of adjacent-pixel conflicting displacements between seam carving iterations. The mean field approximation is used to efficiently solve the problem. In this way, the proposed energy weight can accumulatively calculate the JND information of recent k iterations and passed down to the next iteration for distortion avoidance. The superior performance of the proposed continuous seam carving model, as compared with state-of-art image retargeting approaches in RetargetMe database, was demonstrated experimentally.