建立全對焦影像與各景深場景還原技術

Abstract

在場景中通常會存有多重的目標物，使用傳統的光學或數位相機取像時，受限於鏡頭的光學特性，只能選擇其中一個深度作為成像對焦的依據，因而造成了其他景物的成像模糊，這種現象使得一張照片上普遍存在對焦不佳的部份，對於需要完整記錄景物內容的應用會造成攝影師很大的困擾，特別是當場景內存在物體間深度變化大的情況。 本論文的目的在藉由可控制焦距鏡頭的操作，對取像範圍內的所有物件尋找出正確對焦的參數並加以記錄，利用這些資訊將能允許使用者於事後任意調整對焦參數及重新產生照片，使之能適用於影像還原、編輯等各式的數位影像應用中。本論文內容包括: 1. 透過Intensity Variance v.s. Focus Step Diagram分析 a. Texture Pixel b. Diffused Pixel c. Uniform Pixel 2. 透過水平與垂直 Maximum Variance v.s. Pixel Location Diagram 及 Maximum Focus Step v.s. Pixel Location Diagram 分析Depth Consistency Check 3. 利用景深一致性撿查決定各pixel location的對焦值 4. 找出各景深中的物體範圍 5. 逆向對焦實驗結果

Generally there are more than one object of interest in a scene. The conventional camera can only focus on certain object, resulting in many unfocused objects. This represents a potential problem when the unfocused objects become a valuable in the later usage, especially objects deployed in a wide depth range. In this thesis a sequence of images of the scene is taken with the manual focus function of a digital camera. The goal is to produce an image with all objects under full focusion, which can be used in any postprocessing tasks such as defocusing, editing, etc. The main scope of this thesis include 1. The utilization of the intensity variance vs. focus step diagram to analyze three types of pixels: texture pixel, diffused pixel, and uniform pixel. 2. The utilization of the maximum variance vs. pixel location diagram and the maximum focus step vs. pixel location diagram to perform depth consistency check between pixels. 3. Determination of the correct focus step of each pixel to produce an image with all objects under fusion. 4. Locating the individual objects in the image. 5. The defocusing operation to produce image with an intended depth.