反射係數為主之兩階段光場壓縮法

Abstract

光場成像法是一種近年來廣受注目的影像為主成像架構， 其擁有容易取樣與成像等優點， 缺點在於需要大量的儲存空間； 而現有的光場壓縮法多半是將光場視是為取樣平面上的二維影像陣列， 成像時則將全部的影像逐一還原後再以投射式貼圖的機制顯示在到螢幕上， 如此一來將會需要大量的貼圖記憶體空間來儲存這些影像。 由於光場在成像平面上所取得的影像有對比簡單而且相似的特性， 本論文中便提出一套將光場表示為成像平面上的影像陣列的兩階段壓縮法。 首先量化各影像的頻譜來減少資料， 再將相鄰的影像所對應的係數聚集起來作進一步的編碼。 動態成像時只需要換原對應目前視點的顏色， 並可以利用快取的機制來提升顯示速度。 這個方法可以達到數百倍的壓縮率， 並且在壓縮率與成像速度之間達成一個平衡。

Light field rendering is one of the image-based rendering techniques in computer graphics and has received a great deal of attention due to its simplicity in data establishment and rendering for displaying, though at the cost of enormous requirement on storage space. In this thesis, a two-stage light field compression is proposed for bringing down the space demand, in which light field information are represented in such a way that similarity across data modules can be fully exploited for achieving remarkable compression rate, together with a relative lower cost in the data reconstructing and display rendering phase. The proposed codec achieves compression ratio ranging from 300 to 1000, and is able to decompress data for real time rendering.