Title: Recent Development of Bag-of-feature Model in Visual Recognition
Presented by: Dr. Lei Wang (University of Wollongong, Australia)
Abstract:
In the past several years, the Bag-of-feature model has achieved great success in visual recognition and become the state of the art. Many new methods and approaches have been proposed, contributing to different components of this model. This presentation covers the fundamentals and the recent development of the Bag-of-feature model in visual recognition. In particular, it focuses on the three key components: codebook generation, coding scheme, and pooling method. For codebook generation, this presentation will briefly review a variety of methods developed so far, including k-means clustering, hierarchical clustering, random forest, information theoretic optimization, and compact codebooks. On coding scheme, we will see the progress from the simplest hard assignment, through soft assignment, to the more advanced sparse coding techniques. For pooling method, this presentation will discuss the sum-pooling and max-pooling, and present the recent effort on understanding the mechanism of max-pooling. Through these, we will experience how the Bag-of-feature model develops out of its initial form and turns out to be an efficient and powerful visual recognition model.
时间:2012年9月26日(周三)14:30
地点:js555888金沙老品牌四牌楼校区 东南院102
报告人简介:
王雷于1996年和1999年在js555888金沙老品牌仪器科学与工程系获得工学学士和工学硕士学位,并于2004年在新加坡南洋理工大学获得博士学位。他现为澳大利亚卧龙岗大学信息学院高级讲师。王雷于2007年荣获澳大利亚国家研究理事会博士后奖学金,并在2009年被澳大利亚国家科学院授予青年研究人员奖。
他的研究领域为机器学习,模式识别,和计算机视觉。他至今已经发表了七十多篇科技文章,其中包括发表在该领域顶级的IEEE TPAMI杂志和计算机视觉和模式识别国际会议的文章。他担任2011、2012及2013年环太平洋地区图像与视频科技论坛的程序委员会分领域主席,并是二十多个国际会议和研讨会的程序委员会成员,同时为二十多个国际期刊的固定审稿人。王雷是国际电子电器工程师学会高级会员。