Abstract: Variational method and deep learning method are two mainstream powerful approaches to solve inverse problems in computer vision. To take advantages of advanced optimization algorithms and powerful representation ability of deep neural networks, we propose a novel deep convolutional neural network for image reconstruction. The architecture of this network is inspired by our proposed accelerated extra proximal gradient algorithm with the incorporation of two types of prior-exploiting operations. They are a non-local operation to exploit the inherent non-local self-similarity of the images, and a sparsity-promoting operation to learn the nonlinear transform, under which the solution is sparse. Our experimental results showed that the proposed CNN outperforms several state-of-the-art deep neural networks with similar or even less number of learnable parameters. It also incorporates a non-local operation to exploit the non-local self-similarity of the images and to learn the nonlinear transform, under which the solution is sparse. Our experimental results showed that our network outperforms several state-of-the-art deep networks with similar number or only slightly increased number of learnable parameter.
Time: June 12ed 9:00-10:30
Location: Teaching Building 3, 338
Speaker: Prof.Yunmei Chen
Prof.Yunmei Chen, Distiguished Professor of Mathematics, University of Florida.
Area of Interest: