20160608-18王旗龙:Codebook-free single Gaussian for image classification

报告嘉宾2：王旗龙 （大连理工大学）

报告时间：2016年6月8日（星期三）晚21:00（北京时间）

报告题目：Codebook-free single Gaussian for image classification

主持人：  朱鹏飞（天津大学）

报告摘要：The purpose of image modeling is to capture and represent the inherent information in images, which plays a key role in image classification. The bag-of-features (BoF) model for image classification has been thoroughly studied over the last decade. Different from the widely used BoF which represents images with a pre-trained codebook, an alternative codebook-free image modeling method, which we call codebookless model (CLM), attracts little attention. In this talk, we present two kinds of CLMs where each image is represented with a single Gaussian for effective classification. First, we conduct experiments on eight widely used databases to evaluate the influence factors on our CLM with hand-crafted features. The results show CLM is very competitive to state-of-the-art BoF methods (e.g., Fisher Vector). Following the similar pipeline, we propose a robust approximate infinite dimensional Gaussian (RAID-G) for image representation. We extend RAID-G by using the outputs of deep convolutional neural networks as original features, and propose a new regularized MLE method based on von Neumann divergence for robust estimation of very high dimensional covariance matrix. The RAID-G is evaluated on five material benchmarks and one fine-grained benchmark, achieving much higher classification accuracies than state-of-the-arts. Finally, we give a brief introduction of our most recent work which explores the Lie group structure of Gaussian manifold, a key problem in handling Gaussians.

参考文献：

[1] Qilong Wang, Peihua Li, Wangmeng Zuo, and Lei Zhang. RAID-G: Robust Estimation of Approximate Infinite Dimensional Gaussian with Application to Material Recognition, CVPR 2016 (accepted).

[2] Peihua Li, Qilong Wang, Hui Zeng and Lei Zhang, Local Log-Euclidean Multivariate Gaussian Descriptor and Its Application to Image Classification, IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2016 (in press). 

[3] Qilong Wang, Peihua Li, Wangmeng Zuo, and Lei Zhang. Towards effective codebookless model for image classification. Pattern Recognition 2016 (in press).  

报告人简介：Qilong Wang received the B.E. and M.S. degree in Computer Science and Technology from Heilongjiang University, in 2011 and 2014, respectively. He is currently the second year PhD student at the School of Information and Communication Engineering, Dalian University of Technology. He was a visiting student working with Prof. Lei Zhang for six months, in the Department of Computing, The Hong Kong Polytechnic University. His research interests are primarily on statistical modeling and machine learning with applications to image classification and retrieval. He obtained best 10% paper award in ICIP2015, and as a key member achieved the second place in Alibaba Large-scale Image Search Challenge 2015. He has published 3 scientific articles at top journals including IEEE TPAMI, Pattern Recognition, and 6 papers at top conferences including CVPR, ICCV, and ECCV.

[Slides]