Face Recognition Based on Self-Quotient Image and Biltateral Filter

[[abstract]]在各種生物認證技術中，人臉辨識的應用十分廣泛。在人臉辨識的領域中，大多數的研究僅針對人臉特徵做辨識。很多研究忽略了常見的光照變化問題。因此，本研究結合了特徵辨識與光照辨識。將輸入的人臉圖像先透過小波轉換縮小尺寸，然後藉由自商圖像的技術將人臉圖中的光照變化過濾，並保留過濾掉的光照。為了除去自商圖像處理後的人臉圖的雜訊，本計畫欲透過雙向濾波器來達到此目的。去除雜訊後，將過濾出來的光照變化(OMF_I)與自商圖像(OMF_E)個別輸入到SVM中訓練及測試。如此一來將得到兩個特徵集的辨識率，透過對數轉換與型態學轉換來評估原始圖的光照變化劇烈程度，藉此評估來做為兩個特徵集之辨識率的權重值，並將其結合算出最終辨識率。在進行步驟中，將包含建構系統的流程說明與實驗資料的來源及欲進行的實驗項目等。在系統實作完成後，本計畫欲採用Yale B與ORL這兩種不同性質的人臉資料庫來進行實驗評估。[[abstract]]In a variety of biometric authentication technology, the face recognition is widely used. In the field of face recognition, most of the researches focus on facial feature recognition. Many studies have ignored the normal illumination change. Such identification is not accurate in practice. Therefore, this study combines the features of recognition and light identification. Through this approach makes face recognition to be more robust. In this study, using self quotient image as the main method of human face recognition. The input image is reduced by applying wavelet filter. The self-image after image processing operators and by its light filtered into two sets, respectively OMF_E and OMF_I. Then the core functions of SVM as a classifier. The bilateral filter is used for removing noises while preserving the edge information without destroying the facial characteristics. In this research, the test image databases are Yale B and ORL.[[note]]NSC100-2221-E327-04