Local symmetry features in image processing

Sammanfattning: The extraction of features is necessary for all aspects of image processing and analysis such as classification, segmentation, enhancement and coding. In the course of developing models to describe images, a need arises for description of more complex structures than lines. This need does not reject the importance of line structures but indicates the need to complement and utilize them in a more systematic way.In this thesis, some new methods for extraction of local symmetry features as well as experimental results and applications are presented. The local images are expanded in terms of orthogonal functions with iso-value curves being harmonic functions. Circular, linear, hyperbolic and parabolic structures are studied in particular and some two-step algorithms involving only convolutions are given for detection purposes. Confidence measures with a reliability verified by both theoretical and experimental studies, are proposed. The method is extended to symmetric patterns fulfilling certain general conditions. It is shown that in the general case the resulting algorithms are implementable through the same computing schemes used for detection of linear structures except for a use of different filters.Multidimensional linear symmetry is studied and an application problem in 3-D or in particular, optical flow, and the solution proposed by this general framework is presented. The solution results in a closed form algorithm consisting of two steps, in which spatio-temporal gradient and Gaussian filtering are performed. The result consists of an optical flow estimate minimizing the linear symmetry criterion and a confidence measure based on the minimum error. The frequency band sensitivity of the obtained results is found to be possible to control. Experimental results are presented.