Avaliação não supervisionada do processo de segmentação de imagens utilizadas em geociências

Abstract

With the need of extracting the information contained in satellite images in a quick, efficent and economic way computational image process tecniques are being used more frequently, such as the automatic segmentation. Segmenting an image consists on dividing it in regions acording to a similarity standard, where the pixels which are contained there have the same characteristic, for example, level of gray, texture, that is, the one that best represents the objects on the image. There are lots of examples of segmentary algorithm like the development of areas where the pixels ''grow" and are agglutinated forming regions. To determine which are the best parameters utilized in these segmentary algorithms it is necessary to evaluate the results from the methods used more often, they are the supervized where there is a need of a reference image, considered ideal, giving us a priori knowledge of the regions in study. The unsupervised, where there is not the need of a reference image, make the user save time. Due to the difficulty of obtaining evaluators, for different kinds of images, is proposed the methodology that allows to evaluate images that have vegetation areas, where it will be formed large regions (Crianass), and the one that will evaluate the images with urban areas, where it will be needed more detailing (Cranassir).