植基於影像空間特徵分析之車牌辨識與影像查詢技術

License Plate Recognition and Image Retrieval Techniques Based on Image Spatial Feature Analysis

賴璟皓(Ching-Hao Lai)

指導教授 ： 吳憲珠

國立臺中科技大學/資訊學群/資訊科技與應用研究所/碩士(2006年)

摘要

在本論文中，主要探討二個與影像辨識及查詢之相關主題：車牌辨識 (LPR) 與以影像內容為基礎之影像查詢 (CBIR)。一般來說，車牌辨識之整個流程可以分為三個階段各別進行探討研究，它包含了：車牌偵測、字元切割與字元辨識。在本論文中，針對字元辨識方面，提出了二個以直方圖分析技術為基礎的車牌字元辨識技術。其中第一個方法為四分直方圖（quadratic histogram）分析方法，此方法將欲比對之車牌字元切割為四個部份並針對各個部份建立垂直及水平方向之像素值累積直方圖。在字元比對的過程中，於每一個階段中針對字元各部分之垂直及水平直方圖比對，並在下一個階段中剔除字元樣本資料庫中較為不相似之字元，本方法可以降低直方圖分析比對法在比對過程中的計算複雜度（computing complexity）。經實驗證明，四分直方圖分析方法可以改善直方圖分析法在字元辨識時的效率。
在本論文中，另一個車牌字元辨識方法為三部份邊緣直方圖（tripartite edge histogram）分析方法，此方法將欲比對之字元切割為上、中、下三個部份後，針對各個部分使用八個不同的邊緣遮罩擷取字元的邊緣特徵，並統計各種邊緣出現之機率，以建立邊緣直方圖做為字元比對之特徵。實驗結果顯示本字元辨識方法具有相當不錯之效率及準確性。
另一個研究主題是一個以影像內容為查詢依據之影像查詢技術。在此類的影像查詢技術中，由於所有查詢的依據都以影像內容為主，所以要如何從其影像中擷取出具有區隔能力之特徵，是一個非常重要的關鍵，也是大部分相關研究之研究重點。在本論文中，提出了一個基於Z字型方向掃描方式擷取特徵值之影像查詢技術，在擷取影像特徵值的過程中，針對每一個像素（pixel）及其相鄰的三個像素建立一個 的區塊，並在區塊中使用Z字型方向比較區塊內各像素值的大小關係，以擷取影像中像素值的變化狀況。在本方法中，共歸納出了八種可能出現的狀況，比較各左右相鄰區塊所屬之狀況，藉此建立一個 的共相關性矩陣（co-occurrence matrix；CM）。此矩陣在本方法中稱之為Z字型掃描共相關矩陣 (Z-scanning co-occurrence matrix；ZSCM)。在影像相似度測量的階段中，各影像之Z字型掃描共相關矩陣會被用來作為各影像間之差異度測量的主要評估標準。在本實驗結果中，發現本方法對於紋理及色彩相似之影像具有相當不錯的影像查詢能力。針對另一個使用樣式共相關矩陣(motif co-occurrence matrix；MCM)做為影像查詢特徵之技術與本方法之查詢結果做比較，在不借助簡單色彩直方圖分析技術的狀況下，使用與本方法相同之查詢影像及影像資料庫，本方法可以得到較佳之查詢結果。

關鍵字

車牌辨識 ； 影像查詢

並列摘要

In this thesis, two major research topics about image recognition and retrieval, license plate recognition (LPR) and content-based image retrieval (CBIR) are researched. LPR is usually divided into three stages which include license plate detection, character segmentation and character recognition. For character recognition, two character recognition schemes are proposed, and they are based on the histogram analysis technique. The first proposed scheme is quadratic histogram analysis scheme. It divides the license plate characters into four parts to construct vertical and horizontal histograms by accumulating the pixel values respectively. The character recognizing procedure is divided into four stages. In each stage, the vertical and horizontal histograms of each partial character are mapped to obtain the similarity between characters, and the pattern characters which are less similar as the query character are eliminated in the next stage. Hence, this scheme can reduce the computing complexity of traditional histogram analysis technique in the mapping procedure. The experimental results show that the proposed scheme can improve the efficiency of traditional histogram analysis. Another character recognition scheme is tripartite edge histogram analysis scheme. It divides the license plate characters into top, medium and bottom parts equably and uses eight different edge masks to extract the edge information in each part of characters. The edge information of each partial character is accumulated to construct a partial edge histogram, and the edge histograms of each character are used to be the features in the recognizing procedure. The experimental results show that the proposed scheme can gain more efficiency and accuracy. For the second topic, the CBIR technique uses the content of images to perform the similarity measurement. In order to perform the image retrieval by the content of image, some features must be extracted from images. In this thesis, a Z-scanning feature extracting scheme is proposed for image retrieval. In the feature extracting procedure, each pixel and its three neighbor pixels are used to construct a block. In each block, any neighbor pixels are compared by the Z-scanning order to extract the variation between pixel values. This scheme finds eight different relationships to construct an Z-scanning co-occurrence matrix (ZSCM). At the image similarity measurement, the ZSCMs are the major features of image discrepancy evaluation. The experimental results show that the proposed scheme can gain better retrieval results than the motif co-occurrence matrix (MCM) technique.

並列關鍵字

License plate recognition ； Image retrieval

參考文獻

[1] R. Brunelli and O. Mich, “Image Retrieval by Examples,” IEEE Transactions on Multimedia, Vol. 2, No. 3, 2000, pp. 164-171.

[4] M. I. Chacon and S. A. Zimmerman, “License Plate Location Based on a Dynamic PCNN Scheme,” Proceedings of IEEE International Symposium on Computational Intelligence in Robotics and Automation, Vol. 2, 2003, pp. 1195-1200.

[5] S. L. Chang, L. S. Chen, Y. C. Chung and S. W. Chen, “Automatic License Plate Recognition,” IEEE Transactions on Intelligent Transportation Systems, Vol. 5, 2004, pp. 42-53.

[6] P. Chang and J. Krumm, “Object Recognition with Color Co-occurrence Histograms,” Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Vol. 2, 1999, pp. 498-504.

[7] C. C. Chang and Y. K. Chang, “A Fast Filter for Image Retrieval Based on Color-Spatial Features,” Proceedings of Software Engineering and Multimedia Applications, 2000, pp. 47-51.