基於人因分析之增強人眼觀賞三維影像舒適度演算法

Abstract

現今立體顯示技術已經被廣泛地運用在多媒體應用，在這十年間，已經有許多立體視覺的研究與應用。傳統電影院觀賞立體影像的方法有兩種，分別是利用紅藍眼鏡與偏光眼鏡來區分左右影像，使左右眼個別接收不同的影像。當你的眼睛接收到的存在著視差的影像對，立體的感覺就會被人的大腦所產生。 但有些立體影像在拍攝時並沒有考慮到品質(例如:解析度、控制攝影條件精確度)，所以這種立體影像可能就會讓人感到不舒服。 我們根據觀賞者的對於立體視覺內容的舒適感與偏好建立一個完整的自動化流程調整影像內容。首先我們使用進階的全域搜尋區塊比對演算法與動態規劃演算法估測視差，然後根據人類視覺舒適條件使用深度繪圖法產生新的舒適影像。在深度繪圖法中，我們利用移動拍攝相機的距離調整視差，並且所有立體視覺繪圖的式子可以被近似成一個簡單的式子藉以調整視差。

Stereo display technique is popular for multimedia application now day. Researches and applications on stereo vision have been existed for decades. In conventional stereo theater, one watches movies with “3D Anaglyph Glasses” or “3D Polarizer”. The purpose is to make our eyes accept two different images. When our eyes watch the two images with eyes’ parallax difference, the 3D image will be constructed in our brain automatically. Sometimes, if the 3D contents represented by conventional 3D video format are filmed without quality controlled (e.g. high definition, precise control of the shooting conditions), so they are uncomfortable possibly. We present a complete flow of automation for tuning stereoscopic 3D content based on viewers’ comfort and preferences. 2D image disparities are computed by exhaustive block matching based estimation algorithm and dynamic programming algorithm. Human visual comfort models are applied to analyze the image disparities and guide the depth tuning (shifting/scaling) in order to generate new stereo views by DIBR with desired and comfortable depth perception. The 3D image warping equations for DIBR are introduced. In our system the shift-camera-separation setup is used, and the 3D image warping equations are simplified to a formula which implies horizontal parallax. Real-life image results are shown to demonstrate the effectiveness of our approach.