Optimization of Non-rigid Demons Registration Using Flower Pollination Algorithm

Multimedia technology revolution emerges the improvement of digital image and video processing development. In order to localize and identify different parts or scene of any object in a video frame, video registration is used. This is done according to the fixed reference frame of the same video scene. Demons algorithm can be defined as an image registration technique which is based on the concept of efficient local transformation model. In this paper, image registration was applied on tennis video which was previously divided into multiple image frames. Prior to applying image registration technique, a binning strategy was built which determined the middle frame of the bin (set of frames), and was set as the target frame (middle frame = target frame/fixed reference frame). Afterward, optimization of the demons algorithm-based image registration was done using flower pollination algorithm (FPA). Demon’s registrations’ velocity smoothening kernel parameters were optimized with respect to the correlation efficient which was set as the fitness function. In order to verify the efficiency of the proposed system, the results of current work were compared with the results of optimization using firefly algorithm (FA)-based image registration technique. The comparative study showed that the FPA-based system managed to achieve a superior correlation value of 0.6109 compared to FA-based system's 0.5591. The structural similarity index (SSIM) comparison of both systems also showed that FPA achieved better results with 0.4770 than FA-based system's 0.2736. Additionally, it was also observed that required convergence time was lesser in FPA-based system than FA-based system.