The rapid growth of computer vision applications, such as object detection and identification, has intensified the demand for efficient image segmentation. In underwater image object detection is a challenging job due to the presence of noise. In this manuscript FPGA implementation of two benchmark techniques K-means and K-medoids clustering are carried out for side scan sonar images segmentation. The proposed K-medoids implementation features Verilog modules for random medoid initialization, Manhattan distance-based pixel assignment, and medoid updates, validated by a testbench for one iteration. A K-means clustering architecture is implemented for comparison. The performance evaluation is carried out using segmentation accuracy, FPGA resource utilization, processing latency, and power consumption. System stimulation on KLSG-II side scan sonar image set findings indicate that K-medoids offers robust segmentation, particularly for noisy images, at higher resource costs, while K-means provides faster, resource-efficient processing. These insights guide FPGA-based solutions for real-time image segmentation.