Sualtı görsellerde iyileştirme yöntemleri kalite testi için bir veri seti

Underwater images serve as an invaluable resource for researchers across diverse scientific disciplines. These images provide critical insights into aquatic ecosystems, underwater archaeology, marine biology, geology, and more. However, capturing high-quality underwater images is inherently challenging due to the unique and often unpredictable environmental conditions. Factors such as light absorption, scattering, turbidity, and the presence of particles in the water can severely distort the colors and details of these images, making it difficult to accurately represent the underwater reality. These limitations pose significant challenges, particularly in tasks that require precision, such as underwater photogrammetry, where accurate three-dimensional reconstructions of submerged terrains or objects are essential. Photogrammetry itself is a highly specialized methodology that involves capturing a large number of overlapping images of an object or location from various angles to construct a detailed and accurate three-dimensional model. When applied underwater, photogrammetry becomes even more critical, as it enables researchers to map, document, and analyze objects or environments that are otherwise difficult or impossible to access. Examples include shipwrecks, coral reefs, underwater caves, and submerged archaeological sites. The success of underwater photogrammetry relies heavily on the quality of the images, which must retain sufficient detail and color accuracy despite the challenges posed by the aquatic environment. In recent years, significant research efforts have been dedicated to improving the quality of underwater images through advanced restoration and color correction techniques. These efforts aim to mitigate the effects of light distortion and loss of detail, thereby enhancing the utility of the images for scientific and practical applications. A wide array of approaches has been developed, ranging from traditional mathematical models that address specific aspects of image degradation to state-of-the-art artificial intelligence (AI) and computer vision-based methods. These modern techniques leverage machine learning algorithms and deep neural networks to adaptively correct distortions, enhance image clarity, and restore accurate colors, often surpassing the capabilities of conventional methods. The integration of advanced image processing techniques with underwater research is paving the way for more precise and detailed studies, enabling scientists to better the underwater world with unprecedented accuracy and efficiency. This study is presents an analysis of currently available underwater image datasets for image restoration and enhancement, a thorough analysis of the state-of-the art enhancement methodologies, as well a new dataset of underwater images taken specifically for photogrammetry purposes. This dataset was then used to train and compare currently available methodologies in a different context than what they are designed for, which is a generalist approach to all underwater environments as opposed to presenting a model that is specifically trained for certain types of environments. This study provides a comprehensive analysis of the currently available underwater image datasets that are widely used for image restoration and enhancement tasks. It thoroughly examines the most advanced enhancement methodologies and their capabilities, identifying their strengths and limitations in addressing the challenges posed by underwater environments. Additionally, the study introduces a new dataset specifically designed for photogrammetry applications. This dataset consists of high-resolution underwater images, meticulously captured in controlled conditions to ensure suitability for three-dimensional reconstructions and precise mapping tasks. The newly developed dataset serves as a unique resource for testing the performance of existing enhancement methodologies in a context distinct from their original design objectives. While many current approaches are generalized to accommodate a wide variety of underwater conditions, this study aims to explore their adaptability and effectiveness in a specialized context. By applying these generalist methodologies to the photogrammetry-specific dataset, the research evaluates their capacity to handle unique challenges such as maintaining structural integrity, color accuracy, and clarity in images intended for precise 3D modeling. The results offer valuable insights into the trade-offs between generalist and specialist models for underwater image processing, paving the way for future advancements tailored to specific use cases like photogrammetry. This study also examines the effectiveness of domain-specific model design in contrast to generalized approaches, using the aforementioned dataset for training and evaluation. Our results demonstrate that a smaller, task-oriented model not only achieves a higher PSNR score but also offers significant advantages in terms of implementation simplicity and computational efficiency. These findings support the conclusion that lightweight, use-case-specific models are more suitable for underwater photogrammetry and image enhancement tasks. By focusing on the specific requirements and constraints of underwater imaging, we show that it is possible to outperform larger, more generalized architectures while reducing both the cost and complexity of deployment.