Optimized multi-threshold image segmentation using a hybrid bee-pollinator algorithm

Segmentation is a critical first step in many image-processing applications and is often used during the initial preprocessing phase. Typically, segmentation relies on thresholding. Multi-level thresholding (MLT) in image processing divides an image into various regions based on pixel intensity values. In binary thresholding, pixels are grouped based on a single threshold value, whereas MLT allows multiple intensity levels. One of the challenges in image segmentation is to determine the appropriate threshold. To address this, optimization models are frequently used to identify the optimal threshold values. However, optimization algorithms have strengths and weaknesses, such as the ability to balance exploration and exploitation, the risk of local optima, and issues with fast convergence. Consequently, fine-tuning the optimization model parameters is necessary to improve performance. This work presents a hybrid optimization technique for MLT, integrating an improved Flower Pollination Algorithm (IFP) with the Bee Foraging Algorithm (BFA). The parameters of the IFP were fine-tuned using the BFA. Extensive experiments using standard images were conducted to demonstrate the effectiveness of the proposed hybrid model. Specifically, for the 'Lena' image at a threshold level of 4, the algorithm recorded a Peak Signal to Noise ratio (PSNR) of 26.98 dB, an Structural Similarity Index (SSIM) of 0.416, a Dice coefficient of 0.95, and a Jaccard index of 0.9048, while maintaining a runtime of 15.2 seconds. A comparative analysis with traditional optimization methods demonstrated the superior performance of the hybrid approach across various image inputs.