一般而言,在自然場景中拍攝時,由於相機的有限動態範圍限制,拍攝到的視訊均屬於低動態範圍視訊。為了解決此問題,典型的視訊擷取為使用off-the-shelf相機,在各frame之間改變曝光值來擷取不同的曝光影像,接著各frame經過重建方法生成失去的曝光影像,然後透過合成方法來取得高動態範圍視訊。在本論文中,分為三個主要步驟,首先,利用random sample consensus (RANSAC) 演算法計算相似轉換模型來移除全域運動,然後採用時間雙向相似及連貫靈敏hashing重建方法,基於曝光時間及亮度值之間的關係,來得到較好的重建結果,接著根據計算出的權重,以multiresolution spline based scheme取得合成影像,並透過低動態範圍色調映色調整影像整體亮度,產生似高動態範圍視訊的結果。根據本研究的實驗結果顯示,本論文所提出的研究方法優於比較的四種現有方法。
Generally, captured video is a low dynamic range (LDR) video since camera sensors have a limited dynamic range in a nature scene. To deal with this problem, typically video captured using off-the-shelf camera, namely, captured different exposure LDR images that alternates exposures for every frame, generated missing exposure LDR images at each frame using reconstructed approaches, and produced high dynamic range (HDR) video through fused rule. In the proposed approach, consists of three main steps, random sample consensus (RANSAC) algorithm is used to calculate similarity transform model to remove global motion. Next, temporal bidirectional similarity (TBDS) and coherency sensitive hashing (CSH) are used to obtain better results based on photometric relation between luminance and exposure time. Then, fused images are achieved based on multiresolution spline based scheme. Finally, an HDR-like video is produced through LDR tone mapping. Base on the experimental results obtained in this study, the performance of the proposed approach is better than those of four comparison approaches.