An ameliorated teaching–learning-based optimization algorithm based study of image segmentation for multilevel thresholding using Kapur’s entropy and Otsu’s between class variance

Abstract

In this paper, multi-threshold image segmentation approaches using an improved teaching–learning-based optimization algorithm (DI-TLBO) are presented and the proposed DI-TLBO-based methods obtain satisfactory segmentation results. This work is presented as follows. Firstly, two random numbers are introduced to determine the learning methods of the learner in the teacher phases and the learner phases of DI-TLBO. Randomness of the learning methods further improves global optimization ability of DI-TLBO. Self-feedback learning phase and mutation-crossover phase are also introduced into DI-TLBO algorithm, which makes DI-TLBO achieve better exploration ability. The comparative results of DI-TLBO with other evolutionary algorithms (EAs) on a set of benchmarks functions demonstrate that DI-TLBO acquires better solution accuracy than other EAs. Then the proposed DI-TLBO algorithm is applied to solve multi-level threshold image segmentation problems modeled by Otsu’s between class variance function and Kapur’s entropy function. Experiments comparing DI-TLBO-based methods with other EAs based approaches on standard test images show that DI-TLBO-based methods possess superior performance in terms of both solution accuracy and stability of segmentation results. Finally, the proposed DI-TLBO-based methods are successfully applied in casting X-ray image segmentation for multi-level threshold. Although the defects in high resolution X-ray image (3072×2400) are easy to be ignored and omitted when being detected artificially, all the defects are segmented perfectly using the proposed DI-TLBO-based methods.