Synthetic Video Generation with Camera Motion Patterns to Evaluate Sprite Generation

There is no proper objective method of evaluating sprite generation methods due to absence of the ground truth sprite images. In this paper, we propose several camera motion patterns to generate synthetic videos from original image. Our camera motion patterns include zigzag, spiral, earthquake, and zoom patterns. Subsequently, we apply sprite generation algorithm on the synthetic video. Objective evaluation is performed by comparing original image and generated sprite image based on peak-signal-to-noise ratio (PSNR), size and error on motion parameter estimation. Our results indicate that our picture PSNR, size, and error on motion parameters are good indication of the sprite quality.