Preventing chaining in alpha-trees using gabor filters

Abstract

Hierarchical segmentation using α -trees can suffer from unwanted leakage or chaining effects, which lower segmentation quality by reducing the depth of the hierarchy. In this paper we introduce a new way to prevent the chaining effect of α -trees. It relies on the odd 2-D Gabor filter. A series of clean, noisy, and blurred synthetic images was used to test the ability of improved α -tree to stop the chaining effect obtaining a 99.8% segmentation accuracy, and we compared it with the contrast-based α -tree proposed previously by Soille. Two remote sensing images were also used to test the performance of the methods on natural images. The results showed that both 4-CN and 8-CN odd Gabor filter based α -trees can prevent the chaining effect efficiently.