Objective quality assessment for image retargeting based on hybrid distortion pooled model
Jianxin Lin,Lingling Zhu,Zhibo Chen,Xiaoming Chen
DOI: https://doi.org/10.1109/QoMEX.2015.7148090
2015-01-01
Abstract:With the increasing popularity of mobile devices, there are more and more screens with heterogeneous resolutions. In order to solve the mismatching problem of images displaying on different screens, various image retargeting techniques have been proposed. However, little effective objective quality assessment metric for image retargeting has been proposed. In this paper, we propose an objective image retargeting quality assessment method based on Hybrid Distortion Pooled Model (HDPM) considering image local similarity, content information loss and image structural distortion. The proposed HDPM method measures the retargeted image's local similarity based on matching the similar block by Scale-Invariant Features Transform (SIFT) features and computing the corresponding blocks' similarity by structural similarity (SSIM). Furthermore, the image content information loss in retargeted image, which is regarded as the SIFT feature loss, is taken into account. Besides, we also consider image's structural distortion in the proposed method, which is based on GLCM (Gray-level co-occurrence matrix). To evaluate the effectiveness of the proposed method, extensive experiments have been conducted, and the results show improved consistency between the proposed HDPM method and the corresponding subjective evaluations.