Avaliação de critérios de heterogeneidade baseados em atributos morfológicos para segmentação de imagens por crescimento de regiões

Abstract

This study assesses the impact of using morphological attributes in the formulation of the criterion that rules the region growing process in image segmentation. Therefore, an extension to the multi-resolution segmentation algorithm proposed by Baatz and Schäpe (2000) was proposed and implemented, allowing testing criteria derived from different morphological attributes. The study used a supervised method to measure the segmentation quality numerically. The ideal segmentation result was represented by a set of reference segments delineated manually for three Quickbird-2 image subsets. For each tested criterion, the optimal values of the segmentation algorithm parameters were determined by a stochastic process that aimed at minimizing the discrepancy between the references and the result of each segmentation. Quantitative and qualitative analysis of the results indicated unequivocally that the introduction of morphological attributes in the heterogeneity criterion, which drives the merging of adjacent segments in the region growing process, can result in a substantial improvement in segmentation quality. The paper highlights the importance of adopting morphological attributes suitable for each object class and discusses the selection of such attributes.