Restauração de imagens subaquáticas baseada em adaptação de parâmetros por aprendizado profundo

Abstract

Underwater image restoration faces challenges due to water absorption, scattering, and low visibility. This paper presents a learning-based approach to enhance underwater images. We use a Convolutional Neural Network (CNN) Regression model to learn optimal enhancement parameters from a diverse dataset, allowing the network to generalize across various underwater conditions. Additionally, we apply intensity transformation techniques, including contrast adjustment, histogram equalization, and gamma correction, to improve image quality. Experiments with real-world underwater image datasets show our method achieves high accuracy, demonstrating its robustness and efficiency in restoring underwater images.