Physics aware joint low light and color correction for underwater image enhancement

The underwater images suffer from low illumination, absorption and scattering varies with the wavelength, poor visibility, low contrast and severe color distortion. These degradations have a negative impact on applications for underwater vision like marine exploration, object recognition and autonomous navigation. To overcome these difficulties, a Physics-Aware Joint Low-Light and Color Correction (PA-JLCC) framework to enhance underwater images has been developed. The proposed approach is lightweight, interpretable and does not require extensive training, in contrast with deep learning-based approaches. The framework estimates ambient light level, builds a depth-aware attenuation proxy by using the logarithmic color ratio, and computes a preliminary transmission map from an underwater image formation model, which is further improved by a total variation regularization. The radiance of the scenes is then reconstructed by model-based inversion, enhanced by multi-scale local contrast enhancement and Gray-World color balancing to improve the visual clarity, maintain structural details and restore natural color appearance. Extensive experiments are conducted on synthetic and real-world underwater benchmark datasets to evaluate the visual and quantitative performance of the proposed PA-JLCC framework. Specifically, the proposed approach achieves the reported Underwater Image Quality Measure (UIQM) and Underwater Color Image Quality Evaluation (UCIQE) results, which not only reflect improved perceptual image quality but also improved color fidelity and underwater contrast in various underwater environments. In addition to this, the computational complexity of the proposed algorithm is linear, O(N), where N denotes the number of pixels, making the framework computationally efficient for image enhancement applications. The efficiency and interpretability of the proposed approach make it applicable to underwater image enhancement systems.