Blurred image region detection and segmentation

Abstract

Estimating the blur region of a scene from a single monocular image is a challenging but important research area in modern image processing and computer vision.Early blur identification methods deal with just one type of blur with strong assumptions, which usually limit the wide use in recent computer vision applications. Thus, recent trend in blur estimation has shifted to handling partial blur cases for practical use.Most existing algorithms first classify each patch into either a non-blur or blur region. Thereafter, blur type classification is performed on the blur regions.However, such approaches include potential risk that incorrect blur region identification may affect the following blur-type classification. In this paper, we present a new framework for blur region identification to overcome deficiency of classical methods. We propose a 3-way blur identification method, which devides an image into non-blur, defocus blur, and motion blur regions.To this end, we employ intuitive and powerful features based on specific criteria well-suited for our 3-way classification problem. We also take a coarse-to-fine technique to obtain accurate results.Extensive experiments have been conducted on real images. Our patch-level evaluation results demonstrate that the proposed method remarkably outperforming the recent algorithm both in terms of accuracy and speed. We also provide more subjective and objective evaluation of the proposed method to validate superiority of our algorithm.