基於型變的雙眼影像新視角生成快速演算法

Abstract

本論文提出一個以雙眼立體影像的基於型變的新視角影像生成方法。裸眼立體顯示器需要多視角影像（包含圖片及影片）作為輸入，但現今大多立體相機只能擷取兩個視角（左右眼）的影像。近年來較為流行的新視角影像方法，如深度影像繪圖法，大多很依賴準確的深度圖以取得好結果，然而準確的深度圖目前仍是難以取得的。我們所提出的方法不需要深度圖，並且也不需要使用者的介入。對於生成多視角圖片的部分，首先我們偵測密集且可靠的特徵點。接著我們利用特徵點的對應關係去引導圖片的型變以生成新視角影像，並且在型變的過程中保持立體性質及維持內容結構。基於相同的架構，我們修改了針對圖片的方法使之能處理影片，加強了對於時間軸上的對應關係並且加快生成的速度。相較於深度影像繪圖法，我們提出的方法可以以高效率產生高品質的多視角影像，而且可以免除惱人的參數設定。此方法可以直接將立體相機所拍攝雙眼影像轉換成能在裸眼立體顯示器上撥放的多視角立體影像。

This thesis presents a warping-based novel view synthesis framework for both binocular stereoscopic images and videos. Autostereoscopic displays require multiple views while most stereoscopic cameras can only capture two. Popular novel view synthesis methods, such as depth image based rendering (DIBR), often heavily rely on accurate depth maps, which are still difficult to obtain. The proposed framework requires neither depth maps nor user intervention. To synthesize multi-view images, it first extracts dense and reliable features. Next, feature correspondences guide image warping to synthesize novel views while simultaneously maintaining stereoscopic properties and preserving image structures. Based on the same framework, a modified method for binocular videos are proposed to better maintaining temporal coherence and accelerating the processing speed. Compared to DIBR, the proposed framework produces higher-quality multi-view images and videos more efficiently without tedious parameter tuning. The method can be used to convert stereoscopic images and videos taken by binocular cameras into multi-view images and videos ready to be displayed on autostereoscopic displays.