Polygonal Simplification for Shape Recognition and Representation
Abstract
Shape recognition is a well established problem in the areas of Human Computer Interaction and Computer Vision. The majority of the approaches used to tackle this problem rely on Machine Learning, a process of statistical modeling used to generate patterns and classify data. The contours extracted from images for the purpose of shape recognition often contain extraneous information that complicates the classification process. One such type of noise is caused by baseline shadows, shadows cast from overhead lighting. In addition, these complicated contours pose a problem for systems which store or send full shape information. The reduced shape representation problem asks for the smallest shape representation that retains the maximum area and “closeness” to the original contour. We attempt to eliminate these shadows and simplify the contour information using a computational geometry technique called Polygonal Chain Simplification without significant loss in accuracy. Our results show promise in addressing the baseline shadow problem and show a significant reduction in shape representation.
Description
Keywords
machine larning, computer vision, computational geometry, computer science, human computer interaction, polygonal chain simplification