Detection of Blur and Non-Blur Regions using Frequency-based Multi-level Fusion Transformation and Classification via KNN Matting

In digital images, blur wraps significant information and makes automatic image analysis a challenging task for computer vision algorithms. Hence, accurate blur detection and classification becomes essential to understand the information wrapped up in blurry images. In this paper, we proposed a novel automatic blur and non-blur region detection, and classification technique “Frequency-based Multi-level Fusion Transformation” (FMFT) to detect the unwanted blurry region and classify the blur and non-blur regions using single image processing. Our proposed approach mainly works with frequency sub-bands in patch wise manner, without having any prior information regarding camera configuration, type of blur and intensity of blur. Moreover, the detected Tri-Map from FMFT is further processed to perform blur and non-blur regions classification along with sharp object detection from blurry images using KNN-Matting. The average F1-score of 0.98 signifies the effectiveness of the proposed method in terms of blur detection and classification. Additionally, the proposed method also outperforms existing state-of-the-art techniques.