以像素射線方法開發光場相機之研究

Abstract

近幾年手持式光場相機於影像攝影學上掀起了一場革命性的討論。雖然光場相機的概念早在多前年就已經被提出來討論，但卻受限於當時電腦技術的發展因此無法被廣泛的應用。而在一般光學相機系統中，相機之感測器為一個二維平面，於捕捉現實世界中的四維光場資訊時，所擷取到之二維資訊為整體四維空間場域投影至相機二維感測器平面之結果。換句話說，一般相機之感測器所能紀錄之資訊僅僅為空間光場中之一部分而已，其餘維度之資訊並無法被記錄下來。 本論文目標為計算真實四維光場資訊中之像素射線分佈並擷取此像素射線成像與分析比對。我們採用以物理概念為基礎之渲染系統軟體PBRT(Physical Based Rendering Tracing)來進行光線追跡。初期將會實踐軟體之簡易相機，接著再加以修改相機演算法加入透鏡組來實踐真實相機效果。接下來將此套軟體環境中引入四維光場概念，並對虛擬場景搭配修改後之真實相機來進行光場擷取與呈現。利用此方式所擷取到之光場資料會儲存於光場檔案中，之後會直接對光場檔案進行渲染以及變焦等處理，不用再對整個場景重新進行渲染。最後則將真實光場量測資料導入，置換於PBRT軟體環境中所建立的四維光場中，先會對一張量測資料來做像素射線之光能量擷取與分析，再來利用兩張量測資料來做處理。而本論文會將過往實驗時的貝索光束量測資料與生物組織量測資訊導入。後期利用影像處理方法，對光像素射線之能量分佈圖進行光能量分析，並與真實量測資料來比較。由實驗結果中可看出，透過本論文研究方法，於渲染虛擬場景時可以大幅減少渲染時間至少兩倍以上，另外驗證能夠於PBRT此套軟體環境中達到真實資料之光場光線追跡的目的，並能有效地擷取透過真實資訊後之像素射線光能量分佈呈現於一張相片上。

Recently, the invention of hand-held light field camera induces a revolution in photography. Although the concept of light field camera had been proposed for many years, the limitation of lacking proper computation technology has prevented the light field camera from being widely used. In typical camera systems, the sensor array of a camera is on a two-dimensional plane. To adopt a two dimensional sensor array to capture four dimensional light field data in the world, the data collected represents two dimensional projection of the four dimensional light field data. As the result, the information we get is only part of the whole four dimensional data, other information is simply lost. The main part of this work is to develop the computation approach to calculate the distribution of pixel ray of four-dimensional light field of the real measurement light field data. We use the PBRT (Physical Based Rendering Tracing) software, i.e., the concept of physically based rendering system, to compute the ray tracing necessary to create light-field camera. Firstly, we will use only the simple camera for simplicity and for verifying the system performance. We then add lens system to practice the effect of this modified camera. Secondly, we will introduce the concept of four dimensional light field to the modified camera and capture the light field of virtual scene with the modified camera. With this approach, the captured light field data will be stored in the file of light field. We will then render or zoom this file of light field only, i.e., no need to re-render the entire scene to save time and computation resources. Thirdly, we replace the four dimensional light field by inserting measured real data into PBRT. However, we will capture and analyze the result of pixel ray light energy by using one piece of the measurement data firstly and then using the two measurement data of light field to replace the four dimensional light field. The measurement data used was derived from the Bessel beam data obtained from our team’s previous research results and the newly biological tissue data. Finally, we analyze the pixel ray light energy by image processing and compare the computed result with that the measured data. The result showed virtual scene rendering developed in this thesis can significantly reduce the rendering time needed. The final result clearly demonstrated that the method developed in this thesis, can effectively capture the pixel ray light energy, which can be used to recreated any photo with any specified focal point.