以人工智慧模型修復超穎透鏡影像品質之研究;Research of improving metalens image base on artificial intelligence model

NCU Institutional Repository > 理學院 > 照明與顯示科技研究所 > 博碩士論文 > Item 987654321/92942

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/92942

题名:	以人工智慧模型修復超穎透鏡影像品質之研究;Research of improving metalens image base on artificial intelligence model
作者:	黃晨輔;Huang, Chen-Fu
贡献者:	照明與顯示科技研究所
关键词:	人工智慧模型;超穎透鏡;影像修復
日期:	2023-08-21
上传时间:	2024-09-19 16:32:20 (UTC+8)
出版者:	國立中央大學
摘要:	在當今世界中，光學透鏡在我們的日常生活中佔據的很重要的地位，其廣泛的被運用在各種科技設備中，包括智慧型手機、自動駕駛傳感器和擴增實境(Augmented Reality, AR)及虛擬實境(Virtual Reality, VR)設備中，這些設備都有朝著更薄、更輕的消費電子產品的趨勢，創造了對光學元件不斷微型化的需求。然而，傳統的折射透鏡，如凸透鏡或凹透鏡材料，因為其體積龐大，嚴重阻礙了光學元件的微型化。而近年來，在次波長尺度上的奈米結構發展取得了顯著進展，對超穎表面的關注日益增加。這些奈米結構超穎表面利用集體共振來在奈米尺度上控制電磁波的特性，這一進展為創建微型光學元件展開了可能性。我們對超穎透鏡進行了光學特性的分析，並利用兩個連續的人工智能模型來解決超穎透鏡拍攝的影像中由於超穎透鏡的材料損耗和結構散射所導致的模糊與色偏問題。在人工智能模型這部分，我們分別使用自動編碼器和CodeFormer來校正色偏和重建影像細節。我們透過自動編碼器模型成功解決了所有面部影像的色偏，而CodeFormer模型則可以有效地重建了標準正面臉部、帶有臉部表情的面部細節和側面臉部影像，透過這樣的連續兩個人工智慧模型提升了超穎透鏡在日常生活的應用潛力。;In today’s world, optical lenses play a vital role in our daily lives and are widely used in various technological devices, including smartphones, self-driving sensors, and augmented reality (AR) / virtual reality (VR) equipment. These devices are trending towards thinner and lighter consumer electronics, creating a demand for the continuous miniaturization of optical components. However, traditional refractive lenses, such as convex or concave materials, are bulky and severely hinder the miniaturization of optical components. In recent years, there has been significant progress in the development of nanostructures on sub-wavelength scales, leading to a growing interest in metasurfaces. These nanostructured metasurfaces utilize collective resonances to control the characteristics of electromagnetic waves at the nano-scale, opening up possibilities for the creation of miniature optical components. We conducted an analysis of the optical properties of the metasurface lens and used two sequential AI models to address the blurriness and color cast issues in images captured by the metasurface lens due to material loss and structural scattering of the metasurface lens. In terms of AI models, we used an Autoencoder and CodeFormer to correct color cast and reconstruct image details, respectively. We successfully addressed color cast in all facial images using the Autoencoder model, while the CodeFormer model effectively reconstructed standard frontal faces, facial expressions, and side profile images. Through these two sequential AI models, we have increased the potential for the application of metasurface lenses in daily life.
显示于类别:	[照明與顯示科技研究所 ] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	32	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....