Retrieval of Labeled Shape Classes in Binary Images Using Standardized Granulometric Intervals

Alexandre Y. Harano and Ronaldo F. Hashimoto

Article (PDF) | BibTeX entry

pt-br: Apresentação (PDF) | WVC 2011

ABSTRACT: This work proposes an empirical non-interactive method to identify and retrieve Labeled Shape Classes from binary images. The method bases on granulometries extraction from connected components resized by standardized ratios. The extraction procedure is used to compound an array of representative granulometric intervals on setup phase, as well as to test the connected components on retrieval phase. The use of symmetrical convex structuring elements for granulometry is intended to provide rotation invariance for angles multiples of 90° while the standardized resizing is preconceived to provide scale invariance. The corresponding algorithms are presented, along with their results and proposed extensions.