Blindly Evaluate the Quality of Underwater Images via Multi-perceptual Properties

The quality of underwater images can vary greatly due to the complexity of the underwater environment as well as the limitations of imaging devices. This can have an effect on the practical applications that are used in fields such as scientific research, the modern military, and other fields. As a result, attaining subjective quality assessment to differentiate distinct qualities of underwater photos plays a significant role in guiding subsequent tasks. In this study, an effective reference-free underwater image quality assessment metric is proposed by combining the colorfulness, contrast, sharpness, and high-level semantics cues while taking into account the underwater image degradation effect and human visual perception scheme. Specifically, we employ the low-level perceptual property-based method to characterize the image’s visual quality, and we use deep neural networks to extract the image’s semantic content. SVR is then used to create the quality prediction model by analyzing the relationship between the extracted features and the picture quality. Experiments done on the UWIQA database demonstrate the superiority of the proposed method.