本論文以複雜環境背景的影像為考量對象,提出一套車牌自動辨識的技術。先前已有很多方法被運用於有限制環境的條件下,如車行有固定路線之背景固定的情況、受限制的照度環境等,而我們所提的方法可在影像中偵測任意方位、位置和大小的車牌,更甚者,車牌影像可以是在任意天候及照度情況下取得的,也可以包含錯綜複雜的背景。我們的方法包括兩個主要步驟:車牌定位與車號辨識。在車牌定位部份,利用模糊集合理論和色彩理論,結合車牌的色彩(色調、飽和度及明度)和邊線兩類特徵,以偵測影像中各種類型的車牌。在車號辨識部份,則引進類神經網路學科來實作。實驗結果展示令人滿意的辨識率,進而證明本文所提的技術之可應用性。
In this thesis, we present a technique for automatic recognition of vehicle license plates from images of complex scenes. Many previous approaches were applied to the scene images having settled backgrounds, or being taken under restricted illumination conditions, or both. The proposed approach will be able to detect vehicle plates with discretional orientations, positions, and sizes in images. Furthermore, scene images can be taken under arbitrary weather and illumination conditions and may contain intricate backgrounds. Our approach consists of two major steps: the localization of car plates and the recognition of license numbers. Both the disciplines of fuzzy sets and neural networks are introduced to implement the steps. Experimental results demonstrating satisfactory recognition rates reveal the applicability of the proposed technique.