Efficient Characteristics Vector Extraction Using Auto-seeded Region-Growing

Abstract

Region labeling is an important task in automatic image processing. It consists of assigning information (labels) to each pixel regarding their position, level, etc. and also information regarding the group to which each pixel belongs. This information is useful for many diverse purposes. There are several approaches within this field to perform this task, among these, the region growing implementation has been chosen due to its feature extraction efficiency and flexibility. This approach splits the image into different regions according to different inclusion and exclusion rules that are applied to each pixel. The algorithm proposed is based on an automatic implementation, thus an auto-seeded function has been programmed in order to jump from one region to the adjacent one. Since in real-life images the inclusion is ambiguous, an adaptive implementation has been proposed which allows a pre-defined level of tolerance to gray level variation and, thus, automatically merges regions where the difference is below a specified threshold. The results obtained from synthetic and real-life images are presented in this paper along with a discussion on the results obtained.