A Probabilistic Approach to Restore Images Acquired in Underwater Scenes

Modern imaging devices can capture faithful color and characteristics of natural and man-made scenes. However, there exist conditions in which the light radiated by objects cannot reach the camera's lens or it is naturally degraded. Thus, the resulting captured images suffer from color loss. This article addresses the problem of underwater image restoration by using an optics-based formulation to model the interaction between light and any underwater suspended particle. Our approach uses a factorial Markov random field (FMRF) to reformulate and solve the general nonlinear participating media optical model. This novel formulation also has the particularity of considering attenuation coefficients, beside global light, as to probabilistic latent variables, inferred from a single image. Due to this unique feature, our FMRF methodology for itself is enough to deal with images acquired in underwater scenes. The generality of our optical model makes it applicable in other participating media such as fog or haze, more commonly addressed in the current literature. Results have shown the capabilities to improve the degraded images using our methodology in several scenarios.

Automated summary

This article introduces an optics-based approach for underwater image restoration. By considering attenuation coefficients and global light as latent variables, a factorial Markov random field is used to reformulate and solve the general nonlinear participating media optical model. Experimental results demonstrate the effectiveness of this approach in improving degraded images in various scenarios.