Image analysis using mathematical morphology: Algorithms and architectures

Frank Yeong-Chyang Shih, Purdue University

Abstract

This thesis presents the recently developed image morphology techniques including the algorithms in object feature extraction, recognition, and inspection and the architectures in the implementation of gray scale morphological operations and the decomposition of large structuring elements. The theoretical background and morphological transformations are first introduced. The recursive adaptive thresholding, distance transformation, skeletonization, and reconstruction algorithms are next presented. The recursive adaptive thresholding transforms a gray level image into a set of binary regions. The distance transformation converts a binary image which consists of object (foreground) and non-object (background) pixels into an image where every object pixel has a value corresponding to the minimum distance to its background. The skeleton is extracted from the local maximum of the distance transformation. Next, the decomposition of certain gray scale morphological structuring elements is presented. A non-increasing symmetrical non-linear structuring element can be segmented into a combination of multiple structuring components. This non-linear structuring element can be convex, concave, or a combination of both. A convex structuring element is decomposed into the sequential dilations of segmented linear structuring components. A concave structuring element is decomposed into a selection of a maximum of the results of dilating with each segmented linear structuring component modified with a flat interior region. This allows large structuring elements to be implemented using recursive operations of small structuring components. A threshold decomposition architecture for the gray scale dilation and erosion is next presented. This architecture, using only logic gates, significantly improves speed as well as gives a new theoretical insight into the operations. Application algorithms for industrial parts recognition and inspection are next presented. A morphological edge operator and a gear inspection algorithm using morphological opening and closing are described. Then the feature extraction for corners and circular holes is discussed. A library learning procedure using a hierarchical shape database is developed which leads to the implementation of fast object recognition and location using the shape database. At the end, conclusions and recommendations are made.

Degree

Ph.D.

Advisors

Mitchell, Purdue University.

Subject Area

Electrical engineering

COinS