Research Of Retargeted Image Quality Evaluation Based On Multi-feature Fusion

In order to displaying images on different size device as completely as possible,images need to be retargeted,and it is very important to maintain the image quality at the same time.The content-aware image retargeting methods are based on the content-based smart retargeting,which aims to spread the visual distortion to the relatively unimportant image areas during the retargeting process,thereby reducing the visual error.The difference in the principle of the retargeting algorithm makes the final retargeted effect different.The objective evaluation methods which are consistent with subjective perception can not only assess the retargeting effect,but also promote the development of retargeting technology.However,the correlation between existing objective evaluation criteria and the subjective evaluation is still relatively low.In view of the above situation,this thesis analyzes and quantifies the image content loss and structural deformation caused by the retargeting process,and proposes an objective retargeted image quality evaluation method that integrates multiple features.The features are divided into two categories based on technical characteristics: local retargeting features,and global retargeting features.The main work of the thesis includes:For the local deformation detection,there are two features used: Improved Aspect Ratio Similarity(IARS)and Centralized Deletion Detection(CDD).(1)IARS.Based on Aspect Ratio Similarity(ARS),the improvement of the parameters and formula makes it more suitable for the deformation detection of the retargeted image and the correlation with the subjective perception get higher.(2)CDD.CDD is used to measuring the loss of a large area of the image content and measuring the screen deficiencies of the retargeted image.For the global deformation detection,there are two features proposed: Elastic Registration on B-spline(ERB)and Foreground Object Retention(FOR).(1)ERB.Based on the elastic registration of B-spline function,ERB calculates the structural shape variables between before and after image retargeting in the whole region,and measures the structural deformation in the global image.(2)FOR.FOR can measure the foreground overall retention for retargeted images with clear foreground and background distinction.The proposed objective retargeted image quality evaluation method is obtained by linearly fused the four retargeting features.The simulation experiments are performed on Matlab platform with open access database MIT,NRID(NTHU Retargeting Image Dataset)and CUHK.The experimental results show that the proposed image retargeted quality evaluation algorithm has better correlation with subjective evaluation than the state-of-the-art methods.