2014-08-012024-05-17https://scholars.lib.ntu.edu.tw/handle/123456789/693577摘要：光場攝影在近年廣泛地受到學術界和工業界的關注。不同於傳統的二維影像， 光場攝影技術能夠紀錄空間中各種不同方向的光線，獲得更加豐富的四維資 訊，因此可以得到傳統相機所無法記錄的資訊。但是光場攝影的相關研究與應 用仍有許多待解決的問題，解析度不足即為其一。以目前市場上可取得的Lytro 光場相機為例，雖然其內藏的感光體數量超過1000 萬顆，但是處理過後，最 後能夠得到的影像資料只有約100 萬畫素，遠低於其感光元件的數量。換句話 說，若以像素數量來計算，感光元件所記錄的值有超過90%都被用於換取光線 的角度資訊，使得光場相機即使有大片感光元件，最後得到的影像解析度仍遠 遜於絕大多數的手持裝置相機。若是減少角度資訊來換取解析度，最後得到的 影像會因角度取樣密度不足而產生嚴重瑕疵。因此，如何在空間解析度以及角 度解析之間取得最佳的平衡點是在光場攝影領域必須解決的一個重要課題。本 研究所探討的問題除了具備學術理論價值之外，也兼顧實用價值。 本計畫分三年執行，主要內容包括(1)開發用於四維光場資訊之超解像演算 法，以解決目前渲染光場時解析度不足之問題。(2)開發用於光場角度資訊之 反鋸齒演算法，以解決角度資訊不足時所產生的渲染瑕疵。(3)研究由光場資 訊計算場景深度值與幾何結構之方法。(4)研究光場之空間取樣密度與角度取 樣密度在不同條件與要求下之最佳平衡，並建立相關理論之基礎。<br> Abstract: Light field photography has drawn extensive attention from both academia and industry recently. Being essentially different from traditional two-dimensional photography, light field photography is capable of recording light rays traveling in different directions in space, Thus it can acquire richer four-dimensional information compared to the traditional photography. However, there are still many problems in this emerging field. One of the most serious problems is resolution reduction. Take the commercially available Lytro light field camera as an example. Although the sensor of a Lytro camera is composed of 11 million pixels, the resolution of the final rendered image is merely 1 million. In other words, over 90% of the values recorded by the sensor pixels traded off for angular information. Consequently, although it has more sensor pixels than the camera embedded in smart phones and tablets, the rendered image quality is still much worse than these devices in terms of resolution. If one sacrifices angular information for more spatial resolution, the rendered images suffer the step-edge artifact caused by insufficient angular samples. Therefore, how to strike a balance between the angular information and the spatial resolution is an inevitable problem. In this research plan, we try to solve problems that are not only of academic value, but also essential to make light field photography technique more practical. We propose a three-year research project that contains the following important parts: (1) Design super-resolution technique for four-dimensional discrete light field data, (2) Develop proper anti-aliasing algorithms to solve the step-edge artifact caused by insufficient angular samples, (3) Investigate depth-from-light-field methods to compute the geometric properties of the scene, and (4) Establish a theoretical foundation for optimal trade-off between angular information and spatial resolution under various conditions that may encountered in practice.計算攝影學光場相機相機陣列超解像computational photographylight field cameracamera arraysuper resolution