アダプティッブ領域拡張法

概要

論文の詳細を見る
The segmentation of images Into regions is a fundamental problem in low level computer vision This paper is concerned with the problem of the most popular image segmentation technique of region growing. In the existing region growing algorithms, a predefined, fixed threshold is usually used through out the entire image or on every regular subdivision of the image Because of the diversity of the regional feature homogeneities, there may not exist an optimal threshold to properly segment the complicated image with many an-regular sub-structures. To solve the problem, we introduced an hypothesis test that is taken on a nested multiple-layer neighborhood system to associate every pixel in the image with a grain property, which reflects the homogeneity of the region into which the pixel belongs. Then we can set a local threshold for every pixel in the image to do the region growing.
社団法人電子情報通信学会の論文
1996-03-11