20160713-22 孙剑：Iterative Optimization Algorithms in Image Recovery are Deep.. ...

报告嘉宾：孙剑（西安交通大学）

报告时间：2016年7月13日（星期三）晚21：00（北京时间）

报告题目：Iterative Optimization Algorithms in Image Recovery are Deep Networks?

主持人：孟德宇（西安交通大学）

报告摘要：In this talk, I will show that many iterative optimization algorithms in image restoration and reconstruction can be modeled as deep networks, by which the parameters in iterative algorithm and the corresponding energy function can be discriminatively learned for specific tasks from training images. I will first introduce our proposed discriminative parameter learning approach in Markov random field models [CVPR 2011], which motivates our current research along this direction. Then I will show that the iterative shrinkage in signal processing [IEEE TIP 2015], ADMM (Alternating Direction Method of Multipliers) algorithm optimizing a compressive sensing model can be formulated as deep networks. These deep networks are non-conventional, and achieved state-of-the-art results in image restoration and compressive sensing MRI.

I will also briefly introduce our work on part-based image recognition, deep learning-based image deblur and human organ segmentation. 

参考文献：[1] Jian Sun, Marshall Tappen. Learning Non-Local Range Markov Random Field for Image Restoration. IEEE Conf. Computer Vision and Pattern Recognition, Colorado, USA, 2011.

[2] Jian Sun, Marshall Tappen. Separable Markov Random Field and Its Application in Low Level Vision. IEEE Transactions on Image Processing, Vol. 22, No. 1, Pages:402-408, 2013

[3] Jian Sun, Jian Sun, Zongben Xu. Color Image Denoising via Discriminatively Learned Iterative Shrinkage. IEEE Transactions on Image Processing, 24(11): 4148-4159, 2015.

[4] Jian Sun, Jean Ponce. Learning Discriminative Part Detectors for Image Classification and Cosegmentation. International Conf. Computer Vision (ICCV), Sydney, 2013

[5] Jian Sun and Jean Ponce. Learning Dictionary of Discriminative Part Detectors for Image Categorization and Cosegmentation, International Journal of Computer Vision, DOI:10.1007/s11263-016-0899-0, 2016.

[6] Jian Sun, Wenfei Cao, Zongben Xu, Jean Ponce. Learning a Convolutional Neural Network for Non-uniform Motion Blur Removal. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), Columbus, USA, 2015

[7] Heran Yang, Jian Sun*, Huibin Li, Lisheng Wang, Zongben Xu. Deep Fusion Net for Multi-Atlas Segmentation: Application to Cardiac MR Images (Oral presentation), International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2016

报告人简介：孙剑，西安交通大学数学与统计学院特聘研究员、博士生导师。主要研究方向为图像处理与分析、医学影像分析。2009年获得西安交通大学应用数学博士学位(导师：徐宗本院士)。2005-2008年在微软亚洲研究院计算视觉组从事访问研究(导师：孙剑首席研究员)；2009-2010年在美国中佛罗里达大学（UCF）计算机视觉实验室从事博士后研究（合作导师：Marshall Tappen教授）；2012-2014年在法国巴黎高师和法国国立信息与自动化研究院(INRIA)从事合作研究（合作导师：Jean Ponce教授）。在图像领域国际会议CVPR,ICCV,MICCAI和国际期刊IJCV, IEEE TIP等共发表学术论文25篇。入选教育部新世纪优秀人才计划、西安交通大学青年拔尖人才计划，获得中国工业与应用数学学会优秀青年学者奖。