Development and assessment of novel and accurate methods in shape analysis and computer vision

Abstract

Before image processing and vision techniques can be effectively applied in a truly comprehensive fashion, the involved techniques and implementations have to reach high performance standards. A particularly promising possibility for achieving such a relevant and challenging objective consists in comprehensive and integrated approaches to the typical related problems, including shape representation and its use as subsidy for visual analysis, classification, and synthesis. The inherent advantage in such approaches is that the specific demands and qualities of the processes at each subsequent stage (for instance representation and classification) can be properly identified and complemented. At the same time, it becomes essential to devise and apply formal performance validation and comparative assessment procedures, capable of properly characterizing the potential and short comings of the considered and developed techniques. The current project can be divided into the following two principal parts: (a) development and validation of new representations of visual information,(particularly shapes) and their respective use in analysis, classification and synthesis of 2D, 2.5D and 3D shapes, with special emphasis on scale-space approaches based on differential geometry, partial differential equations, signal processing, and skeletonization; and (b) developments and applications in the areas of tunelling microscopy (2.5D images), face recognition, and industrial parts identification. Additional research topics also to be considered in the planned developments include techniques for multiscale image processing and analysis (such as wavelets and mathematical morphology), digital video processing and motion analysis. (AU)