Utilisation of borderline continuity and convexity in segmentation of complex head and shoulders images

Abstract

This Master of Science Thesis describes an image segmentation method and its application, namely the procedure for removing complex background from head and shoulders images. The method's applicability to extraction of facial borderlines is shoulders images. The developed scheme is a partial result of a process of developing a versatile set of image processing algorithms which can be used e.g. in image analysis. The method extracts a borderline, which has a number of specific characteristics from a source intensity image. The principle feature of the borderline is that it encloses a certain reference point that is passed to the method as an input parameter. The other preferred characteristics of the borderline are defined in terms of a cost-function that consists of a number of different error functions. Optimal borderline is obtained by minimising the cost-function value. /Kir10