Abordagem wavelet para detecção de cantos em formas

Abstract

Shape analysis plays an important role in computer vision and image processing applications and shape representation is the first step towards it. The current dissertation discusses a set of techniques available in the literature in order to generate an adequate shape representation based on corners. The proposed approach provides an non-decimated wavelet decomposition of the represented contour by curvature and angulation signals in order to represent the shape using the most relevant contour points without requiring any user intervention. The adopted methodology presents promising results and it uses a set of tools, with little highlight in the literature, and it eliminates the redundant shape features in order to accomplish the suitable representation of the original shape. A novel evaluation measure, concerning the reconstruction error, is also presented to confirm that the proposed algorithm achieves its goal and outperforms other corner detectors used for tests and comparison.