Uma abordagem multi-escala para segmentação de imagens

AUTOR(ES)
DATA DE PUBLICAÇÃO

1999

RESUMO

ln this work we consider the problem of image segmentation by means of a multiscale representation. This multiscale representation is based on a recently proposed morphological scale-space theory, the Multiscale Morphological Dilation-Erosion - MMDE, which associated to the watershed transform, reduces monotonicly the number of extrema of an image and, consequently, the number of its segmented regions. This method has two basic problems concerning image segmentation: the spatial shifting of the watershed lines throughout the scales and the di:fficulty to characterize the set of the image extrema across these different scales. The first problem is considered here by means of a homotopic modification of the original image. The MMDE approach states that the position and the amplitude of the extrema in the original and transformed images do not change across scales. This property allows us to use a set of these extrema, present at a certain scale, as marker in a homotopic modification and segmentation of the original image. Also, we consider an analysis of the way the image extrema merge across scales and introduce a new morphological scale-space in which the monotonic reduction of the image extrema is given by an idempotent operation. For this scale-space, We consider some monotonic-preserving merging criteria, taking into account only local information, which can be used to prevent significant image extrema from merging and to define better sets of markers for segmentation

ASSUNTO(S)

processamento de imagens

ACESSO AO ARTIGO

Documentos Relacionados