学术报告第201218期

报告题目：Random  Forest for Image Annotation or Fast Semantic  Nearest Neighbour Search

报告人：Guoping Qiu  (邱国平教授)

报告时间：2012年12月19日下午14:30---15:20

报告地点：南一楼中311室

Abstract:

This talk presents a novel method for automatic image annotation (which  can also be understood as a fast semantic nearest neighbour search method). We  use the tags contained in the training images as the supervising information to  guide the generation of random trees, thus making the retrieved nearest neighbor  images not only visually alike but also semantically related. Different from  conventional decision tree methods, which fuse the information contained at each  leaf node individually, our method treats the random forest as a whole, and  introduces the new concepts of semantic nearest neighbors (SNN) and semantic  similarity measure (SSM). We introduce a method to annotate an image from the  tags of its SNN based on SSM and have developed a novel learning to rank  algorithm to systematically assign the optimal tags to the image. The new  technique is intrinsically scalable and fast, and we will present experimental  results to demonstrate that it is competitive to state of the art image  annotation methods. (Contents of this talk has appeared as “Hao Fu, Qian  Zhang, Guoping Qiu: Random Forest for Image Annotation. ECCV (6) 2012:  86-99”)