객체 추적을 활용한 비디오 인페인팅용 사용자 기반 마스크 생성 기법

초록

When unintended objects such as passersby appear in media, traditional methods like covering the object are used to removethem, which can lead to unnatural results. Instead of covering objects, video inpainting can offer a better alternative. However,video inpainting needs to generate masks for every frame, which can be quite costly. In this paper, we propose user-based objecttracking method for video inpainting to address this problem. Our framework combines object tracking and segmentation toautomatically generate masks for object removal in video inpainting. Since it tracks user-specified objects, it enables the selectiveremoval of desired objects, even when multiple objects are present. The proposed method achieves better results compared to usingonly video segmentation for mask generation and reduces per-frame execution time by half.

키워드