A survey on automatic image semantic segmentation techniques

PhD Qualifying Examination


Title: "A survey on automatic image semantic segmentation techniques"

by

Mr. Honghui Zhang


Abstract:

emantic segmentation, assigning each pixel in an image to one of several pre-de 
ned semantic categories, has many application of high practical value. In this 
survey, some state-of-the-art semantic segmentation techniques are 
reviewed,including the learning based methods and the label transfer based 
methods. In the learning based methods, semantic segmentation is treated as a 
supervised classi cation problem.They usually train a statistical model by 
using some given training data fi rst, like the widely graphical model, and 
then segment new images with the trained model. The label transfer based method 
works with a totally di erent way. The semantic segmentation problem is reduced 
to match the test image to an existing set of images with annotation, and 
transfer the annotation from these annotated images to the test image. Together 
with the advantages and disadvantages of each class of methods, some future 
directions in semantic segmentation will be discussed in this survey.


Date:                   Monday, 30 August 2010

Time:                   2:30pm - 4:30pm

Venue:                  Room 3501
                         lifts 25/26

Committee Members:      Prof. Long Quan (Supervisor)
                         Dr. Huamin Qu (Chairperson)
                         Dr. Pedro Sander
 			Dr. Chiew-Lan Tai


**** ALL are Welcome ****