一個利用幾何性質由數位影像中偵測直線,圓,及橢圓的研究

Abstract

霍氏轉換法是一個由數位影像中偵測參數性曲線的有效方法。然而，霍氏 轉換法需要大量的計算時間和空間用來產生所有可能的參數組合並將它們 累積在參數空間的相對位置上。許多霍氏轉換法的改進版本利用邊緣點的 梯度資訊來分解參數空間以達到減低時間及空間需要量的目的。但是這些 方法仍然受數位影像中的雜訊及空間量化問題的困擾。為了解決這些問題 ，在這邊我們提出了幾個利用直線、圓及橢圓整體的幾何特性來偵測這些 特徵的方法。這些偵測法在沒有用到邊緣點梯度資訊的情況下明顯的降低 了所需產生參數組合的數量。實驗結果顯示了這些特徵偵測法的正確性及 有效性，並且這些方法也被用到地籍圖及機械工件圖的辨識上。 The Hough transform (HT) is a robust method for detecting analytically described curves. However, the use of the HT needs a lot of computing time and storage for evaluating all possible parameters and accumulating them in the corresponding entries of the parameter space. In order to reduce the requirements of computing time and storage, many modified versions of the HT have been proposed to decompose the parameteration into several small stages via edge orientation information. However, these methods suffer from noises and quantization error. To treat these disadvantages, here, we propose several methods for detecting lines, circles and ellipses via their global geometric properties. These detectors significantly reduce the time required to evaluate possible parameters without using edge orientation information. Experimental results are also given to show the correctness and effectiveness of the proposed methods. In addition, it is shown that these methods can be applied to land-register map and machine-part drawing recognition.