Patch-based partial motion blurred segmentation

Abstract

Motion blur has a significant impact on image recognition. Segmentation of motion-blurred regions contributes to further identification or classification. Most existed segmentation algorithms are always universal for partially blur images, but not especially for motion-blurred ones. This paper proposes a particular algorithm aiming at motion-blurred region segmentation. Firstly, motion regions are segmented by a patch-based preprocessing. Then, the blurriness of motion regions is measured by a defined function to detect local blurred areas. Empirical thresholds are recommended according to the experimental results. The experimental results show that the motion-blurred regions can be segmented more accurately, and the speed almost doubles other algorithms. Thus we propose a more accurate and efficient segmentation method, especially for partial motion-blurred images.

Keywords

• Local motion-blurred
• Image segmentation
• Patch-based segmentation