Fast video registration method for video quality assessment

In this paper, we propose a block-based video registration method that can be used for various applications such as objective video quality assessment. Instead of using the entire frames of a reference video sequence and a processed video sequence, we use a limited number of frames which are selected under a certain criterion and the selected frames are divided into a number of sub-blocks. Then, we select a small number of sub-blocks which have large spatial gradients. In order to find such sub-blocks, a spatial filtering is applied to the sub-blocks of the selected frames and we use energy of filtered subblocks to measure spatial gradient. The proposed registration method is fast and experimental results show that the method provides accurate registrations for a wide range of videos.