Utility-Oriented Quality Assessment of Sonar Image Super-Resolution

Sonar sensors are widely used for underwater tasks because of their advantages of working in turbid environment and having a long working range. However, there is a trade-off between working range and image resolution in practical applications. For improving the quality of the acquired acoustic images, image super-resolution (SR) has recently drawn considerable attention. Due to the inconsistent characteristics between natural images and sonar images, the existing image quality assessment (IQA) methods for natural scene images do not well provide constructive suggestions for super-resolution of sonar images. To solve this problem, we first established a super-resolution sonar image (SRSI) database, where the database was built around its usefulness of image for marine tasks. Then, a no-reference IQA specifically designed for SRSIs based on a brain-inspired target recognition mechanism was proposed, which is accomplished in a hierarchically organized ventral visual pathway. Experimental results show that our database has certain reliability and that the proposed metric outperforms the state-of-the-art IQA methods.