International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 05 | May 2024
p-ISSN: 2395-0072
www.irjet.net
ENHANCING UNDERWATER IMAGES WITH REINFORCEMENT LEARNING P. Niranjan Reddy Professor Computer Science and Engineering. Kakatiya Institute of Technology and Science. Warangal, India
Anvitha Kandi Computer Science and Engineering. Kakatiya Institute of Technology and Science. Warangal, India
Mohammed Abdul Sameer Computer Science and Engineering Kakatiya Institute of Technology and Science. Warangal, India
Soniya Chalamalla Computer Science and Engineering Kakatiya Institute of Technology and Science. Warangal, India
Sai Kumar Reddy Rikkala Computer Science and Engineering Kakatiya Institute of Technology and Science. Warangal, India
--------------------------------------------------------------------***--------------------------------------------------------------------Abstract—This large-scale effort aims to improve images by using reinforcement learning approaches to overcome the difficulties related to underwater imagery. Due to the special challenges presented by underwater environments—such as low contrast, color distortion, and limited visibility— traditional image processing techniques struggle to produce acceptable results. This work suggests a novel method for adaptively improving underwater photos that makes use of reinforcement learning algorithms. The study starts with a thorough examination of the components that contribute to underwater image deterioration and an investigation of current improvement techniques. Deep neural networks are then used in a reinforcement learning framework to determine the best improvement policies. Through training on a wide range of underwater photo datasets, the model learns to automatically modify contrast, brightness, and color correction settings. The performance of the suggested model is evaluated using a range of metrics, such as visual comparisons with the state-of-the-art improvement strategies and an evaluation of image quality. The outcomes show how well the reinforcement learning strategy works to dramatically increase underwater image visibility and quality. Experiments carried out in various settings and with diverse illumination conditions are also used to evaluate how well the model adapts to various underwater situations. This study introduces a data-driven, learning-based method to improve image quality, which advances the field of underwater imaging. The model can adapt to a variety of underwater conditions thanks to the application of reinforcement learning, which also automates the enhancement process. The results of this work may find use in marine biological studies, environmental monitoring, and underwater surveillance, where high-quality photography is needed
absorption phenomena. Capturing high-quality images in such conditions is crucial for various applications, including marine biology [3], oceanography, underwater archaeology[2], and surveillance. According to research literature, humans experience between 70% and 80% of their environment through visual information [1]. Traditional image enhancement techniques often struggle to address the complex and dynamic nature of underwater scenes, prompting the exploration of novel approaches such as reinforcement learning (RL). Reinforcement learning, a branch of machine learning, has garnered significant attention in recent years for its ability to learn optimal decision-making policies through interaction with an environment. In the context of underwater image enhancement, RL offers a promising framework for automatically improving image quality by learning to adaptively adjust parameters or operations based on feedback from the environment or human experts. This research paper aims to explore the application of reinforcement learning techniques to the task of underwater image enhancement. By formulating the enhancement process as a sequential decision-making problem, RL algorithms can learn to make intelligent adjustments to image parameters such as contrast, brightness, and colour balance to maximize visual quality and clarity. Moreover, RL-based approaches have the potential to adapt to varying underwater conditions, such as changes in water turbidity, illumination, and scene complexity, making them robust and versatile for realworld deployment. By leveraging the power of reinforcement learning, we aim to advance the state-ofthe-art in underwater image enhancement.
Keywords——Double Deep Q network, Markov decision process (MDP), reinforcement learning, underwater image enhancement.
A. Motivation Many obstacles arise from underwater imaging, making it more difficult to collect and process visual data in aquatic settings. One of the main challenges is light attenuation, which causes less contrast and visibility in underwater photos as light is absorbed and scattered as it passes through water. Water stains and suspended particles can
I. INTRODUCTION Underwater environments present unique challenges for imaging systems, characterized by poor visibility, light attenuation, and colour distortion caused by scattering and
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 1678