在本論文中,我們利用光電相關濾波器來作人臉識別的研究。為了使得偵測面上目標物有一樣的相關輸出峰值,我們提出了MACE-NOJTC 的架構。此架構中,我們使用了 Lagrange multipliers technique來設計所需的理想濾波器函數,當限制所有訓練影像輸出相關峰值為某一特定值的條件下,將所有訓練影像的平均輸出相關能量最小化。我們所得到的理想濾波器函數使用在MACE-NOJTC 的架構上,此系統也同時具有旋轉及位移的不變性。 結合了光與電的優點,在俯仰、旋轉、等環境下仍很成功地辨識出正確影像。利用類似的技術,我們可用在指紋辨識糸統、印鑒辨識系統上,甚至應用於軍事雷達影像偵測上。
In this thesis, we investigate the face pattern recognition using hybrid electro-optical correlation filters. To detect target images with equal output correlation peaks, the minimum average correlation energy nonzero order joint transform correlator (MACE-NOJTC) is introduced. The design is carried out using the Lagrange multipliers technique to minimize the average correlation energy for all training set images while constraining the correlation peak value to a user-specified constant. An ideal reference function for the MACE-NOJTC is derived. Correlation results are presented. It is observed that the system has the ability to recognize the rotated image and to locate the image position. Combining advantages of electronics and optics the correct image is successfully detected under noisy environment. Using the similar technique, we apply to fingerprint recognition system, the Chinese seal recognition, and synthetic aperture radar (SAR) images detection system now.