結合共變異矩陣及自相關不變性於二維物件辨識

Using Covariance matrix and Autocorrelation Invariance for 2D-object Recognition

翁宏源(Hong-Yuan Weng)

指導教授 ： 田方治

國立臺北科技大學/管理學院/工業工程與管理系所/碩士(2005年)

摘要

隨著生產技術不斷地創新改進下，不僅大幅縮短產品產出時間，產出數量亦隨之暴增。面對此趨勢下，為解決生產技術所帶來之效應，產品品質檢驗技術勢必要獲得有效提升，加快檢驗速度及準確性，才足以因應此問題。本研究目的在運用機器視覺技術，結合共變異矩陣之特徵值（Eigenvalue）及自相關係數之不變性（Invariance property），建立一套有效的二維物件之視覺辨識系統。本研究經由兩階段特性擷取（Feature extraction），強化特徵值之能力，改善以往較難克服之物件變化辨識問題。研究內容主要探討二維圖形之不變性，針對在不同旋轉角度（Rotation）、縮放比例（Scale），及隨機位移（Random translation）之二維圖形下，將二維圖形先作二值化，藉由型態學(Morphology)擷取物件邊界點(Boundary points)之座標值，利用共變異矩陣(Covariance matrix)之理論求得物件邊界之特徵值，透過抽樣取一固定數量之特徵值，計算其自相關係數（Autocorrelation coefficient），所得之自相關係數向量與標準樣本之自相關係數向量進行相似性量測（Minimum distance measure）及灰關聯度（Grey relation grade）之分析，另外再應用倒傳遞網路（Back propagation network）學習標準樣板之自相關係數向量並分類，進而判定其歸屬。依據實驗結果，觀察目前所測試100張標準樣本中，灰關聯度分析辨識率可達96.8％（最小距離法為95.7%），倒傳遞網路更可高達99.5％（最小距離法為97.2%）辨識率之成效，顯示分類效果良好。

關鍵字

不變性物件辨識 ； 共變異矩陣 ； 自相關係數 ； 灰色理論 ； 倒傳遞網路

並列摘要

This study proposes an effective recognition system of two-dimensional objects by incorporating the eigenvalues of covariance matrix with invariance of autocorrelation coefficient. The proposed method first uses mean threshold method to derive the image. Then, a morphologic operation is adopted to extract the boundary of a digital object. Two-stage feature extract is designed in this study. First, the boundary of object is represented by calculating the eigenvalues of the covariance matrix according to the region of support. Secondly, the sampled eigenvalues of the covariance matrix is transformed using autocorrelation coefficients to maintain the invariant property. Finally, three supervised classification methods, Grey relation measure, Minimum distance measure, and Back propagation network, were tested in this study. One hundred standard patterns with 10 different digitizations are used for system validation. Experimental results show that the proposed method performs excellently with 99.5% recognition rate in using Back propagation network.

並列關鍵字

Invariant object recognition ； Covariance matrix ； Autocorrelation Coefficient ； Grey relational theory ； Back propagation network

參考文獻

[9] Gonzalez, R. C., and Woods, R. E., Digital image processing, 2nd Edition, Prentice Hall, 2003.

[10]Makridakis, S., and Wheelwright, S. C., Forecasting Methods and application, New York:John Wiley ＆ Sons, 1978.

[13]Sharma, S., Applied Multivariate Techniques, New York:John Wiley and Sons, 1996.

[14]Young, T. Y., and Fu, K.S., Handbook of pattern recognition and Image processing, New York :Academic Press, 1986.

[15]Aihara, N., Iwasa, H., Yokoya, N., and Takemura, H., “Memory-based self- localization using omnidirectional images,” Pattern Recognition, vol:2, 1998, pp.1044-1046.