Image Retargeting Assessment Based on Salient Region Similarity

Jun Wang,Jia Bei,Tongwei Ren,Yan Liu,Bowei Xu

DOI: https://doi.org/10.1145/2382336.2382376

2012-01-01

Abstract:The prosperity of image retargeting technique leads to the significant need of effective approaches for assessing different image retargeting methods. In this paper, we propose a novel automatic approach providing human perception based image retargeting assessment by measuring salient region similarity between the original image and the target image. First, the salient regions in the original image and the target image are matched using gradient, color and direction features. Then, the similarity between the salient regions are measured based on feature distances and region saliency. Based on salient region similarity, the quality of target image is assessed with two criteria derived from real user requirements in retargeting, important content retainment and visual artifact reduction, and the overall scores of the target image quality are finally calculated by integrating assessment results on different criteria. Experiments demonstrate the effectiveness of the proposed approach.