深度學習技術檢測面部肌膚暨擬真彩妝系統開發

Development of Deep Learning-Based Facial Skin Detection System and Photorealistic Makeup Application

本研究計畫旨在結合深度學習、影像預處理技術及影像辨識演算法，開發一套針對人體面部膚況檢測並模擬上色的技術。研究的核心目的和重點是期望開發出整合「色彩校正與結合各類型深度學習框架」功能的膚質檢測APP，且透過嵌入結合White Patch色彩白平衡演算法於影像處理技術中，確保鏡頭捕捉的照片色彩能夠忠實反映實際人臉膚色。此次研究之一重點，則在於利用不同深度學習框架，進行皮膚上的問題檢測，例如透過YOLO來檢測雀斑、痤瘡以及利用Keras作為架構來進行迴歸預測任務，進行皮膚年齡預測。同時，為提供逼真的彩妝體驗，更透過運用dlib工具庫精準定位使用者的五官位置，使用者能夠在自己的照片上模擬不同的彩妝效果。本研究的創新之處，在於針對影像進行色彩校正處理，尤其是運用白平衡技術，以提升照片的真實度，進而提高皮膚分析和彩妝模擬的準確性及實用性。計畫研究的最終目的，為降低使用者在提升面部美感時所需付出之成本。期望此研究能夠快速且有效地找到使用者自身面部肌膚概況，節省尋求醫療協助時所花費成本；透過擬真之模擬彩妝技術，讓使用者快速明確了解自身合適及喜好之妝容，降低在選擇化妝品時可能付出的額外成本；在商業方面，期望市場上的可見度能夠被提升，並吸引可能的潛在廠商「共同合作」─進行進階優化手持行動裝置軟體之開發。

Th i s r e s e a r c h a ims t o d e v e l o p a comprehensive system that integrates various techniques/ algorithms of deep learning, image preprocessing and image recognition for detecting and simulating color application on human facial skin conditions. The core objective of this study is to create an app for skin texture detection that incorporates color correction and leverages various deep learning frameworks. The integration of the White-Patch color balance algorithm in image processing ensures that the captured photographic colors accurately reflect the actual facial complexion. This research primarily focuses on using different deep learning frameworks for skin issue detection, such as employing YOLO for identifying freckles and acne and utilizing Keras for regression prediction tasks to estimate skin age. Additionally, the project offers a realistic makeup experience by precisely locating facial features using the dlib library, enabling users to simulate various makeup effects on their photos. The innovation of this study lies in its approach to color correction in images, particularly through white balance technology, to enhance photo realism and, consequently, the accuracy and practicality of skin analysis and makeup simulation. The ultimate goal of the research is to reduce the cost burden on users seeking to enhance their facial aesthetics. This study aims to rapidly and efficiently delineate individual facial skin conditions, minimizing the financial outlay of seeking medical consultation. By employing photorealistic makeup simulation technology, users can swiftly ascertain their preferred and suitable makeup styles, reducing the additional costs associated with selecting cosmetics. On the commercial front, the study aims to increase market visibility and attract potential collaborators for advanced optimization and development of the app.