Year: 2007
Type: conference-paper
Abstract: The choices of gray-level image feature matching criteria and window sizes are considered.Features, as pixels enclosed in windows around interest points, are matched by operating on statistical measures where a match is declared in accordance to the result of a decision process.The decision error should be minimized as well as a small size window used in order to minimize the computational complexity.In this work, the maximum discrimination criteria is applied in the decision for a guaranteed bound on the matching error.The window size is determined by applying the principle of nonparametric signtest, that satisfies the requirement for an adequate representation of the probability distribution function needed in conducting a match.The effectiveness of the proposed approach is verified by matching features in image sequences captured by a mobile robot, to the features detected and stored in the first frame.
Authors: N. M. Kwok, Mark G. Carmichael, Q. P. Ha, Kailong Tan
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W1202783975