Order-Invariant Toboggan Algorithm for Image Segmentation

Abstract

The toboggan and watershed algorithms are two classic algorithms for image segmentation. Both of them are based on the concept that images can be segmented into homogeneous regions by partitioning the gradient images along the ridges. The watershed algorithm has been referred more frequently in the literature than the toboggan algorithm, partially because the watershed algorithm is conceptually easier in classifying the pixels in the flat regions or on the ridges. In this paper, we present an order-invariant toboggan algorithm, which can classify the pixels in the flat regions and on the ridges. Also, a modified morphological grayscale reconstruction is introduced to preprocess the gradient image so that the over-segmentation problem can be alleviated. This paper demonstrates that the toboggan algorithm is faster than the watershed algorithm both in theoretical complexity and in runtime practice.