應用特徵抽取法的影像自動鑲嵌系統

Abstract

本論文提出一個可使影像鑲嵌(mosaic)過程自動化的特徵抽取方法。此法可在有重疊的影像中找到可靠的對應點。影像中的線、角落常被用來做為特徵，但以線及角落做為影像鑲嵌的特徵點，有可能會使接線的不連續性更加的明顯。本論文找出的特徵將是在一個區域中相對比較特殊的區塊，並且使影像鑲嵌有較佳的視覺效果。 在相機移動及視角變化不是很劇烈的情況下，此法的步驟如下，首先預估出影像中重疊的區域，再從該區域中，抽取出數個特徵。由取出來的特徵驗證初始預估之重疊區域是否在可容許的誤差範圍之內。在估出之重疊區域內，利用所提的特徵抽取方法，找出數個特徵配對，在這些特徵中，找尋最適合的特徵配對。 本論文所提出之特徵抽取方法，在以特徵為基礎的鑲嵌系統中(feature-based mosaic system) ，此法可以提供可靠的特徵配對使鑲嵌結果有較佳的視覺效果。在不需特徵的鑲嵌系統中(featureless mosaic system)也可以提供數組可靠的特徵配對做為初始值，以避免在求最佳解的過程中掉入局部最小值。由實驗結果可以看出，利用所提出的架構，選取出來的數組可靠特徵配對可用於影像鑲嵌技術並且有較佳的視覺效果。

A feature extraction methodology is used to achieve an automatic mosaic system. Reliable features set are found in the overlapping region of two images. Edges and corners are usually considered as features in images but they may not suitable for a mosaic system. A basic assumption in this thesis is that the movement of camera is not drastic. The procedure of proposed feature extraction can be divided into three steps. First, estimate the overlapping region in image sequences which are tried to be combined. Second, use the proposed feature detector to extract significant features in the overlapping area. Then, find the corresponding features in each image. Finally, corresponding feature pairs that have the minimum difference in the neighboring columns are selected for a mosaic system. The proposed method can provide reliable corresponding features in a feature-based mosaic system. The corresponding feature set can be used in a featureless mosaic system to be an initial guess of transformation matrix such that falling into the local minimum can be avoided during the optimization procedure. Experiments are reported that the performance of proposed method. The experimental results show that the proposed method can extract reliably corresponding pairs for the application of video mosaic.