Noisy Sonar Image Segmentation using Reptile Search Algorithm

Abstract

Due to the low energy attenuation of an acoustic wave in water, the side-scan sonar imaging technique is popularly used for underwater exploration. The images collected in this process contain a high amount of noise, which poses a challenge to accurately detecting underwater objects. In this paper, the de-noising of such images is carried out through a non-local means filtering algorithm. The obtained denoised images are further segmented to effectively determine the object, shadow, and background. The segmentation task is formulated as a clustering problem, and a recently reported nature-inspired algorithm known as Reptile Search Algorithm (RSA) is used. The RSA is based on the hunting behavior of crocodiles in a specific region. The Davies-Bouldin index is used as the fitness function to perform the clustering. The performance of the proposed method is evaluated on four plane and four-ship images collected from the benchmark KLSG-II dataset. The obtained results are compared with the image segmentation performed by particle swarm optimization and genetic algorithm. Comparative results reveal that the proposed RSA-based model obtained better results in de-noising and effectively segmenting the eight images.