Gaussian normalization of morphological opening distributions for analyzing texture defects

In this thesis, we develop new morphological criteria called normalized distributions and apply them to analyze texture defects. Texture is universally regarded as a fundamental surface property and is important in industrial quality control and automated inspection applications. Defects in material result in variations in texture properties which may then be analyzed through the use of image analysis techniques.; Mathematical morphology is a non-linear image processing technique that provides a set of tools for analyzing the geometries of image textures. In particular, morphological opening distribution, which is analogous to a sifting operation, has been recognized as a useful texture analysis tool. The performance of an opening distribution is governed by the selection of an appropriate structuring element, which plays the role of a specific sieve in the sifting operation. However, it is found that a single structuring element is not adequate in detecting multiple types of defects. It resulted in the use of a battery of structuring elements, leading to increased computational cost, or in studies to develop an optimal structuring element.; In this thesis, we propose a different approach that does not use a battery of structuring elements, but incorporates a-priori knowledge about particle size distributions into opening distributions. The theory of morphological opening operation is explored in greater detail to study the interaction of structuring elements and image textures culminating in normalized distributions. A Gaussian particle distribution model is used to develop normalized distributions for linear structuring elements and the ability of the normalized distributions to overcome several problems with raw opening distributions is demonstrated with several examples.; The normalized distributions lead to the development of a unique algorithm that does not require segmentation for texture defect analysis. The ability to segment an image and therefore locate the defect are traded-off against an increased capability to detect multiple types of defects. The texture defect analysis algorithm is applied to the problem of pavement distress assessment to demonstrate its ability to detect, classify and measure texture defects.; Pavement distress assessment involves detecting, and measuring the type, extent and severity of several types of defects on pavement surfaces. Pavement structures are composite materials with particles (aggregates) bound together by asphalt or Portland Cement Concrete mix (binder), which results in a characteristic textured surface. Defects such as cracks result in regions of no texture, while pot-holes and other surface defects result in the appearance of secondary textures. The morphological algorithm, using normalized distributions, has been tested on over 400 pavement images and the results show a success rate of almost 95%. (Abstract shortened by UMI.)...