Multi-Seed Region Growing Algorithm for Medical Image Segmentation

Abstract

We present a heuristic approach to segment an image into multiple regions for subsequent feature extraction. The algorithm is based on region growing and allows parallel implementation by employing multiple seeds, that independently grow a region until all pixels of the image have been assigned. Seeds are homogeneously dispersed in pixel space and the growth of regions is controlled by prioritizing neighboring pixels via a bucket queue. The heuristic is based on histograms that are built up during growth to derive binary images for each seed. These binary images are weighted by additive image fusion. A simple preprocessing technique is applied to tune the algorithm’s outcome. We explain how input parameters influence the algorithm’s outcome and how practical solutions can be obtained.