OABC: An Otsu-Artificial Bee Colony Multilevel Image Threshold Optimization for Liver Tumor Segmentation

Abstract

Grey scale or color image conversion involves thresholding technique which can be either at single or multiple threshold levels. Resultant images have reduced pixel level classes. As level of thresholding required increases, the process becomes complex involving exhaustive search and thereby increasing computational resources and time complexity. Meta-heuristic optimization algorithms show promise to reduce time-complexity, especially the Artificial Bee Colony (ABC) algorithm mimicking honeybee foraging activities. It provides optimal results when sampling a large solution space. Since threshold-based image segmentation is a multidimensional discreet optimization problem, for fitness evaluation of threshold level candidates, Otsu’s thresholding method has been widely employed. In this paper, proposed is an Otsu-ABC multilevel image thresholding model for CT images using MATLAB for Liver tumor segmentation. The developed model is evaluated for increased dimensionality and computation time. It is also benchmarked against other commonly used algorithms. Results demonstrate that the proposed model is immune to dimensionality increase and outperforms all other models with regards to quality of images segmented and computation time proving that the model is ideal for optimizing multilevel image thresholding problems involving large thresholds for complex computer vision problems in medical diagnostics, biometrics, surveillance, satellite imagery and gaming.