Title: Fine-Grained Image Classification Based on Discriminant Region Mining
Speaker: Li Haojie
Time: 10: 00-11: 30, Monday, October 28, 2019
Venue: Conference Room 601, Administration Building
Biography:
Li Haojie, Professor, Ph.D. supervisor, and Deputy Dean of School of International Information and Software, Dalian University of Technology. He obtained his Bachelor’s Degree from Nankai University in 1996 and Doctoral Degree from the Institute of Computing Technology, Chinese Academy of Sciences in 2007. He was a Postdoctoral Researcher at the National University of Singapore from 2007 to 2009 and joined the Dalian University of Technology in January 2010. His main research interests are Multimedia Information Retrieval, and Computer Vision. He has published more than 100 papers in top international journals such as IEEE TCSVT, IEEE TIP, IEEE TMM and important international conferences such as ACM MM, ICCV, and IJCAI. He has been awarded the Best Paper Award of ACM ICIMCS (2011), HHME (2013), and ICME (2017) Finalist Award. Also, He ranked the 1st in The Star Challenge International Multimedia Search Engine Competition (2008). In recent years, he has led 1 National Natural Science Foundation of China (Key Program), 3 National Natural Science Foundation of China(General Program), 2 basic Scientific Research Projects of National Defense (including 1 cooperation), and more than 10 projects of Enterprise Cooperation and Provincial and Ministerial Level, including International Cooperation, Scientific Research Foundation for Returned Scholars, Ministry of Education of China. The developed software systems have been widely used in Power, Postal, Sports Training, Intelligent Manufacturing, and other industries.
Abstract:
Fine-grained classification aims to classify the images (cars, dogs, flowers, birds, etc.), belonging to the same basic category, into a more detailed subcategory. It is a very popular research topic in the fields of Computer Vision and Pattern Recognition in recent years and is widely used in various fields (such as Species Classification of Ecology, Commodity Classification of Automatic Sorting, Garbage Classification of Environmental Protection, etc.). Because the differences between subcategories are subtle and usually exist in the local areas, how to efficiently select highly differentiated local areas and how to optimize their characteristics is the focus of the current research. This report will introduce our exploration in these two aspects. Firstly, I will introduce two relationship-guided differential region learning methods from the perspective of local regional correlation mining. Secondly, a characterization optimization method based on the discriminant prior network model and autoregressive model is introduced. Finally, the research trends of fine-grained image classification are discussed.
Organizer: School of Artificial Intelligence, Jilin University
