Using Multiscale Structural Fusion for Sonar Image Quality Evaluation

Sonar imaging has broad and critical applications in marine technologies since it is able to solve the problem that the optical system cannot operate effectively in turbid water. During the process of underwater acoustic channel transmission, sonar images are easy to degrade and thus seriously influence the resultant feature extraction and target recognition. To that end, we devise a new SOnar Image Quality Evaluation (SOIQE) method in this paper for better monitoring and guaranteeing the visual quality of sonar images. Guided by highlighting the important task objects, our proposed SOIQE method makes a pixel-wise fusion of structures extracted at multiple scales. Experimental results on the newly established sonar image quality database show that our SOIQE approach is superior to the state-of-the-art image quality models in terms of implement efficiency and evaluation performance.