SUBPIXEL ACCURATE SEGMENTATION OF SMALL IMAGES USING LEVEL CURVES

Abstract

This paper presents an algorithm for the segmentation of small images which combines edge-based and colour-based approaches to image segmentation. The idea is to trace continuous lines of equal intensity through an image. The starting points for these level curves are determined in an edge detection step. In a postprocessing step the gradient information along a level curve is analysed to assert that it always follows object boundaries. The segmentation algorithm works with subpixel accuracy and unites the good adaption to local image structures from the edge-based and the continuousness of segment borders from the color-based approach.