In Multisensor image registration, the objective is to develop algorithms to register images from different sensors and/or at different times. The plan for the first year is to focus on implementation of the basic contour matching, the elastic matching approach, and evaluation of the methods prototyped using a graphical image processing user interface tool. The research work related to registration focused on extending the method of Li, Manjunath, and Mitra ("A Contour Based approach to multi-sensor image registration,"). The method was implemented in the C programming language and ported to the EXPLORER interface on a silicon graphics computer. Some preliminary testing of this method compared with some standard methods was done at UCSB, and will be further explored on the dataset provided by Professor Goodchild's group in the Geography department. After the two images are registered it is advantageous to combine them so as to present the best resolution of the two images in each part of the overlapped region. For this the group is investigating a wavelet based approach to image data fusion.
For the UCSB/Alexandria digital library project, we are investigating the use of wavelets for image data storage, browsing, and content based retrieval. The group is currently investigating the use of different orthogonal wavelet transforms for storage and browsing. Their research on content based retrieval is currently focused on texture pattern retrieval. They have developed feature extraction algorithms using Gabor filters, and in collaboration with the database researchers, are working on developing efficient indexing structures.
The Department of Electrical and Computer Engineering can be reached on the web at http://www.ece.ucsb.edu//default.html