基於超輕型立體顯微鏡之自動影像校正

傳統的光學手術顯微鏡是雙眼倚靠目鏡觀看立體畫面，新一代的數位手術顯微鏡則是採用感光元件將立體影像呈現在3D 螢幕上。在鏡頭機構設計上，我們必須依照不同放大倍率與工作距離利用步進馬達推動整組鏡頭的傾斜角度以獲取最佳的對位角度，這方面容易有齒輪背隙與硬體磨損問題，造成傾斜角度非我們所期待的位置，導致立體影像產生錯誤。針對上述問題，我們提出利用軟體計算兩顆鏡頭影像的誤差資訊回饋給機構重新調整兩顆鏡頭的傾斜角度，並修正硬體誤差造成的畫面水平位移情況，方法是找出視訊中匹配的區塊並對位，藉此計算對位的位移距離，來快速提供機構校正鏡頭角度所需的數據。特徵點匹配可以有效地找出準確的對位位置，然而其計算複雜且維度高、速度慢是其最大的缺點。因此在本論文中，我們利用中值門檻位元圖（Median Threshold Bitmap）減少計算相似度所需的時間，並利用位元運算（Bitwise）的方式計算相似度，最後在匹配流程中進一步應用影像金字塔（Image Pyramid）減少對位所需要的時間。實驗結果顯示，所提方法可即時的計算出影像對位數據，進而達到即時雙鏡頭傾斜角度與畫面水平校正的目的。

關鍵字

立體顯微鏡； MTB ；影像校正

並列摘要

One of the most common problems of the two-lens digital stereo surgical microscope is in maintaining proper alignment and consistency between the images it produces, because the two lenses often become misaligned or shifted during the transportation or abnormal usage. The traditional ways of solving this problem cannot overcome asynchrony between lenses. Therefore, we propose a software solution to automatically perform real-time video matching from the two lens systems, and then compute the displacement of the video frames and subsequently compute the transformation to align the two lens systems. Video frames registration by feature point matching is reported to have high accuracy, but the features are matched using high-dimensional local descriptors which make the computations very complex and time-consuming. In order to construct a real-time video frame registration method with efficient computation and good matching performance, we first apply a median threshold bitmap (MTB) which has good matching performance with fast computation, as well as retaining important information about the image. We then implement the proposed feature matching method using MTB and image pyramid. Experimental results show that the proposed method has good registration performance, and is fast and suitable for a real-time two lens correction system.

並列關鍵字

Digital Stereo Surgical Microscope ； Median Threshold Bitmap ； Image Calibration

國際替代計量

基於超輕型立體顯微鏡之自動影像校正

全文下載

主題瀏覽