Author ORCID Identifier
0000-0002-0162-1453
Biosketch
Mr. Senthil Anand Narayanasamy is currently working as Assistant Professor in School of Computing at SASTRA Deemed to be University, Thanjavur – 613 401, India. He is pursuing research in the area of Image Processing, Mr. Senthil Anand Narayanasamy obtained his M.C.A., in Bharathidasan University, Trichy and M.Tech., (Computer Science and Engineering) at SASTRA Deemed to be University, Thanjavur. He did his Bachelor’s degree in Computer Science at Bharathidasan University, Trichy. His areas of specialization are Image Processing, Machine Learning and Signal Processing.
Date of Award
24-7-2025
Document Type
Thesis
School
School of Computing
Programme
Ph.D.-Doctoral of Philosophy
First Advisor
Dr.K.Rajkumar
Keywords
Focused Super-Resolution Generative Adversarial Networks, Side-Aware Boundary Localization, Color Compensation, Color Correction, Image Enhancement, Leaf in Wind Optimization
Abstract
Image enhancement is an essential process in numerous fields, including industrial inspection, medical imaging, remote sensing, and photography, as it improves image quality for accurate analysis and interpretation. Among the advanced image enhancement techniques, Focused Super Resolution (FSR) with Self-Attention Single Candidate Optimizer-based Generative Adversarial Networks (GANs) is specifically designed for weld defect detection, while Advanced Image Enhancement through Multi-scale Color Correction and Contrast Stretching using Leaf in Wind Optimization focuses on enhancing the overall visual quality of images. Although both approaches aim to improve image quality, they differ significantly in their objectives and application areas. The FSR method concentrates on recovering fine structural details from low-resolution images for industrial inspection, whereas the latter technique enhances color accuracy, contrast, and visual clarity for a broad range of imaging applications.
The Focused Super Resolution (FSR) technique employs a Self-Attention mechanism within a Generative Adversarial Network (GAN) framework to reconstruct high-resolution images from low-resolution inputs containing subtle weld defects. By integrating a Single Candidate Optimizer, the method effectively identifies important image regions, enhances feature extraction, suppresses noise, and improves the visibility of minute defects. This capability is particularly valuable in automated weld inspection systems, where precise detection of cracks, pores, and other structural imperfections is essential for maintaining industrial quality standards. The enhanced image resolution significantly improves the reliability and accuracy of defect identification during quality control processes.
In contrast, the Multi-scale Color Correction and Contrast Stretching technique emphasizes improving the overall appearance of images rather than detecting specific defects. It utilizes the Leaf in Wind Optimization algorithm to perform adaptive color correction at multiple scales, ensuring natural color reproduction under varying illumination conditions. Furthermore, contrast stretching enhances the visibility of low-intensity regions, allowing hidden details to become more prominent while preserving image quality. This method dynamically adjusts enhancement parameters based on image characteristics, making it highly suitable for applications such as photography, remote sensing, medical imaging, and other scenarios where visual clarity and color consistency are essential.
Both techniques provide effective image enhancement but serve different purposes depending on application requirements. The FSR-based GAN approach is primarily intended for industrial environments, where extracting fine details from low-resolution weld images is crucial for automated defect detection and quality assurance. Conversely, the Multi-scale Color Correction and Contrast Stretching method is a general-purpose enhancement technique that improves brightness, contrast, and color fidelity to produce visually appealing images across diverse domains. Therefore, the selection of an appropriate enhancement technique depends on whether the objective is precise defect detection or overall image quality improvement.
Low-light images frequently suffer from low brightness, poor contrast, high noise, and limited visibility, making image analysis challenging. To address these issues, an effective low-light image enhancement framework first converts the input image into the HSV (Hue, Saturation, Value) color space, where the Value (V) component undergoes multi-scale decomposition using the Sharpening-Smoothing Image Filter (SSIF). The approximation component obtained from the decomposition is enhanced using Contrast-Limited Adaptive Histogram Equalization (CLAHE) to improve brightness and local contrast while minimizing noise amplification. Finally, the enhanced approximation component is combined with the preserved detail components to produce an output image with superior brightness, enhanced contrast, reduced noise, and improved visibility, making the method suitable for surveillance, medical imaging, autonomous driving, and other low-light imaging applications.
Recommended Citation
N, Senthil Anand Mr, "Multi-Scale Color Correction and Contrast Enhancement via Leaf in Wind Optimization for Improved Weld Defect Detection in Non- Destructive Testing" (2025). Theses and Dissertations. 194.
https://knowledgeconnect.sastra.edu/theses/194