A Segmentation Framework for Acoustic Sidescan Sonar Images Using Improved Smallest of Constant False Alarm Rate and MAP-MRF

Segmentation of sidescan sonar image is a significant issue in underwater object detection and recognition. However, most prior methods only consider segmentation accuracy, ignoring false alarm rate, which plays a vital role in object detection and recognition. In this paper, a robust and accurate segmentation framework for sidescan sonar image is proposed, which balances a preferred tradeoff between accuracy and false alarm rate. The proposed method integrates an improved Smallest Of Constant False Alarm Rate (SO-CFAR) algorithm and a Maximum A Posteriori probability and Markov Random Field model (MAP-MRF). The part of innovations segments acoustical highlight region accurately while preserving edge features, which can make segmentation results obtain preferred false alarm rate. After that, MAP-MRF is employed for overcoming drawbacks associated with higher threshold value in continuous acoustical highlight areas. Besides, to better deal with intensity inhomogeneity, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is incorporated into this method, which can locate Region Of Interest (ROI) in sonar images as well as improve segmentation effect. Experimental and comparative results on actual side-scan sonar images demonstrate that our method provides superior denoising, precision, and robustness performance.