影像處理應用之特定多功能影像操作研究

Abstract

現代影像處理面臨多樣且複雜的挑戰，包括光照變化、雜訊干擾與影像模糊等問題。傳統方法往往針對單一任務設計，並假設理想化環境，導致其在實際應用中缺乏通用性與適應性。本研究旨在系統性分析各類影像處理技術及其融合應用，探討在不同環境條件下的性能表現。透過對深度學習方法與傳統演算法的比較與整合，協助推動開發高適應性的影像處理框架，進一步挖掘各技術於多樣場景中的潛在優勢與應用價值。 第一章說明本研究的動機與主要貢獻。第二章聚焦於結構保護式影像平滑技術，探討此類方法如何有效保留影像中的邊緣與結構細節，並評估其在正常光源與低光環境下的應用表現。第三章重點分析邊緣檢測技術，比較基於深度學習與基於傳統演算法的能量計算方法，觀察其在正常與受干擾場景中的表現。第四章則探討不同頻域轉換方法在影像修復中的應用，重點比較離散餘弦轉換（DCT）與傅立葉轉換（FFT）在處理去噪、降雨、霧霾與低光等場景中的表現。實驗結果顯示，DCT方法在提升影像修復品質方面具備明顯優勢。第五章介紹其他先進的影像修復技術，並針對降雨、霧霾與低光等常見退化問題進行評估。此章亦比較這些方法與本研究所提出之基於FFT並結合增益曲線補償的影像復原技術，在不同場景（如白天與夜晚）下的修復效果差異。 本研究的主要貢獻在於探討多種影像處理技術的融合應用，驗證其在多樣場景中的有效性，並為未來影像處理技術的發展提供新的視角與實務參考。

Modern image processing faces numerous challenges such as illumination changes, noise, and blur. Traditional methods, often tailored for specific tasks under ideal conditions, struggle with adaptability in real-world settings. This study systematically analyzes a range of image processing techniques and their combinations, assessing their effectiveness across varied environments. By comparing and integrating deep learning methods with classical algorithms, the research aims to advance flexible image processing frameworks and highlight each technique’s strengths and practical value in diverse scenarios.

Chapter 1 presents the motivation and main contributions of this study. Chapter 2 focuses on structure-preserving image smoothing techniques, exploring how these methods effectively retain edges and structural details in images, and evaluates their performance under both normal and low-light conditions. Chapter 3 provides an in-depth analysis of edge detection techniques, comparing deep learning-based methods (such as EDTER and RCF) with traditional algorithmic approaches based on energy computation. The chapter observes their performance in normal and noise-affected environments and analyzes the fusion of image smoothing and edge detection methods. Chapter 4 examines the application of different frequency domain transformation methods in image restoration, with a particular focus on comparing Discrete Cosine Transform (DCT) and Fast Fourier Transform (FFT) in scenarios such as denoising, rain removal, fog removal, and low-light enhancement. Experimental results demonstrate that the DCT method has a significant advantage in improving the quality of image restoration. Chapter 5 presents other advanced image restoration techniques and evaluates their performance in handling common degradations such as rain, haze, and low-light conditions. It also compares these methods with the proposed FFT-based image restoration approach using gain curve compensation, focusing on their restoration performance across different scenarios, including both daytime and nighttime conditions.

In summary, the main contribution of this study lies in the exploration of the fusion of various image processing techniques, validating their effectiveness in diverse scenarios, and providing new perspectives and practical references for the future development of image processing technologies.