An Improved Flower Pollination Optimizer Algorithm for Multilevel Image Thresholding

Multilevel image thresholding is an important technique for image processing. However, the computational complexity of multilevel image thresholding grows exponentially with the increase in the number of thresholds when using the exhaustive searching method. To address this problem, a plenty of heuristic algorithms are applied to search the optimal thresholds. In this paper, an improved flower pollination algorithm (IFPA) using Tsallis entropy as its objective function is presented to find the optimal multilevel thresholding. In the IFPA, three modifications are utilized to enhance the flower pollination algorithm (FPA). First, an adaptive switch probability method is used to balance the local and global pollination. Second, a new local pollination strategy is adopted to avoid the population falling into local optimum. Third, an crossover and selection operations are applied to the FPA which can increase the diversity of the population, then enhancing the performance of the FPA. Subsequently, three different algorithms such as FPA, GSA and DE are introduced to compare with the IFPA in the experiments. The experimental results demonstrated that the IFPA can search out the optimal thresholds effectively, accurately and can obtain the best image segmentation quality.