區域二元化圖樣 (LBP) 主要是在一個預先決定的特定區塊範圍內以中間像素的灰階值為基礎,對區塊中的其它像素進行門檻化運算。區域二元化圖樣對雜訊很敏感,為了解決這個問題,有人提出一個三值化運算,稱為區域三元化圖樣 (LTP)。延展式區域三元化圖樣 (ELTP) 沒有使用固定的門檻值,其門檻值是由區塊範圍像素的局部統計量來取得,以此自適性的門檻值來提高對抗雜訊的能力。本論文提出一個新的自適性運算子,稱為全區變數自適性區域三元化圖樣 (GVALTP),其門檻值是由整個影像的像素統計後自動決定,此新方法的優點是使運算更快速。由對兩個臉部圖像庫的臉部辨識實驗顯示,我們的技術與其他演算法相比可達到相當或稍好的辨識率,而且所提方法的抗雜訊能力與延展式區域三元化圖樣相當,但卻有比其更快速的運算機制。
The operator of local binary pattern (LBP) mainly thresholds pixels in a predetermined window based on the gray value of the central pixel of that window. The LBP is quite sensitive to noise. To deal with this problem, a 3-valued operator called local ternary pattern (LTP) was proposed. The operator of extended local ternary pattern (ELTP) does not use a fixed threshold rather its threshold is determined by the local statistics of the pixels in the window. With this adaptive threshold, the noise-resistant capability is improved. In this thesis, we introduce a new adaptive operator called global variable adaptive local ternary pattern (GVALTP), where the local pattern threshold is determined automatically based on global statistics of the entire image. The main advantage of this new approach is to enable a faster calculation mechanism. Experiments conducted on two face image datasets show that our technique achieves comparable or slightly better recognition performance as compared with other algorithms. For noise-resistant capability, the proposed approach achieves comparable performance with the ELTP but runs much faster.