Performance Evaluation Approach for Image Mosaicing Algorithm

In order to evaluate performance of any image mosaic algorithm objectively, a complete evaluation indicators system is constructed in this paper based on reference panoramas or non-reference panoramas. The involved indicators reflect human visual effect, image registration accuracy and image fusion effect. Then the formula of registration error (RE) is improved from TRE. Meanwhile, a blind evaluation approach related on the indicators system is proposed to avoid any manual interventions and to keep the computational simplicity. Exhaustive tests show that the proposed approach can reflect performance of image mosaic algorithm comprehensively and the result of experiment is consistent with subjective evaluation.