Role of Resolution in Noisy Pattern Matching

Abstract

Natural variability and limited number of library instances influence the matching performance and robustness of automatic pattern analysis methods. A data representation method that utilizes the local structures from the original data with a focus to effectively use the full resolution of data vector is presented. Using plant and face database, the issue of resolution, limitations in the number of library instances, and variability were addressed. Interestingly, the method shows an automatic matching improvement of over 22% under noisy and difficult matching condition.