Multilevel thresholding is a widely used image segmentation method. In order to improve the efficiency of image thresholding, this paper proposes a Fuzzy Improved Hunter Prey Optimization (FIHPO). In response to the drawbacks of Hunter Prey Optimization (HPO) being prone to local optima, low search efficiency, and poor convergence speed, this paper first uses chaotic tent mapping and elite reverse learning to improve the population initialization of HPO, increasing the diversity of the population. Then, an adaptive weight strategy is adopted to enhance the convergence speed of the HPO. Finally, a Gaussian mutation strategy is used to improve the global optimal search ability of the HPO. In segmentation experiments, fuzzy entropy was used as the FIHPO’s objective function, and median aggregation was used to complete the segmentation of natural and plant images. By comparing with six algorithms, namely Fuzzy Improved Coyote Optimization Algorithm (FICOA), Improved Artistic Bee Colony Using Sine Cosine Algorithm (ABCSCA), Modified Whale Optimization Algorithm (MWOA), HPO, Modified Ant Lion Optimizer (MALO), and Reptile Search Algorithm (RSA), FIHPO can achieve better segmentation efficiency and visual results.