Title: DeFusionNET: Defocus Blur Detection via Recurrently Fusing and Refining Discriminative Multi-scale Deep Features
Authors: Chang Tang, Xinwang Liu, Xiao Zheng, Wanqing Li, Jian Xiong, Lizhe Wang, Albert Zomaya,Antonella Longo
Source: IEEE Transactions on Pattern Analysis and Machine Intelligenc
Published: August 2020
DOI: 10.1109/TPAMI.2020.3014629
Link: https://ieeexplore.ieee.org/document/9161280/
Abstract:
Albeit great success has been achieved in image defocus blur detection, there are still several unsolved challenges, e.g., interference of background clutter, scale sensitivity and missing of boundary details. To deal with these issues, we propose a deep neural network which recurrently fuses and refines multi-scale deep features (DeFusionNet) for defocus blur detection. We first fuse features from different layers of FCN as shallow features and semantic features, respectively. Then ,the fused shallow features are propagated to deep layers for refining the details of detected defocus blur regions, and the fused semantic features are propagated to shallow layers to assist in better locating blur regions. The fusion and refinement are performed recurrently. In order to narrow the gap between different feature levels, we embed a feature adaptation module before feature propagating to exploit complementary information and reduce contradictory response of different layers. Since different feature channels are with different extents of discrimination blur detection, we design a channel attention module to select discriminative features for feature refinement. Finally, the output of each layer at last recurrent step are fused to obtain the final result. We collect a new dataset consists of various challenging images and their pixel-wise annotations for promoting further study. Experiments on commonly used and our newly collected datasets are conducted to demonstrate efficacy and efficiency of DeFusionNet.
