Recently, the latest research result of Prof. Nong Sang’s team “Blind Image Deblurring via Deep Discriminative Priors” was accepted by international top academic journal International Journal of Computer Vision (IJCV). IJCV is one of the top academic journals in the field of computer vision with 11.541 impact factor.

The paper proposed a blind image deblurring algorithm based on discriminative priors, which overcame the problem of poor generalization performance of prior constraints in existing deblurring algorithms. In the research, a binary classifier is trained through convolutional neural networks, which the input is an image and the output is a scalar representing the probability that the input image is blurred. The algorithm embeds the network into the coarse-to-fine MAP framework as a prior constraint to constrain the potential clear image, thus this algorithm can estimate the motion fuzzy kernel. By testing the data of the motion-blurred image, it shows that the algorithm proposed in the paper has clearer boundaries, finer details and fewer artifacts compared with current image deblurring algorithm. Meanwhile, through changing training data, the paper extended the proposed discriminative priors to image defogging algorithm and achieved good recovery effect.

Previously, the paper “Learning a Discriminative Prior for Blind Image Deblurring” was once accepted by international top conference IEEE Conference on Computer Vision and Pattern Recognition (CVPR) and posted. It was a joint work by Prof. Nong Sang and his doctoral student Lerenhan Li, Associate Prof. Changxin Gao, Prof. Jinshan Pan from Nanjing University of Technology, Doctor Wei-Sheng Lai and Professor Ming-Hsuan Yang from University of California at Merced.