應用區域對比增強於不均勻光源下之人臉辨識

Abstract

近幾年來，由於安全上的需求，所以利用人臉來進行身份辨識的應用越來越廣泛，在許多從事人臉辨識的研究的文獻中，常利用人臉影像擷取出來的特徵，來分辨出不同的人。然而在實際的應用上，常常會因為環境中光源的不均勻照射，使得同一張人臉會有很大的不同，因而導致人臉的辨識率大幅下降，為了提昇辨識效能，我們提出一個區域對比增強的方法，可以有效的解決人臉辨識在不同光源下的改變。

本篇論文提出的人臉辨識的演算法，則是在辨識前對影像做離散餘弦轉換，取出人臉影像的低頻部份，有效降低影像的維度，因此在辨識的時間上也會相對的減少，最後交給支持向量機（SVM），來決定辨識的結果。本論文測試的人臉資料庫為Yale_B，經使用支持向量機的辨識率可達99.13%，在已發表的論文中是辨識較好的方法之一。

In recent years, many face recognition algorithms have been developed for surveillance systems and promising results have been reported in specific environments. The human face recognition highly relies on extracted stable features from input images. In practical application environments, however, the direction of the illuminant is uncontrollable and it will result in unstable feature extraction. For remedying the problems caused by non-uniform light sources, illumination compensation is necessary. In this thesis, we propose a local contrast enhancement approach to reduce the effect of non-uniform light sources, and integrate it with a face recognition system. Through the process of local contrast enhancement, the facture extraction based on digital cosine transformation (DCT) becomes more reliable. The adopted classification kernel is support vector machines (SVM) which has been shown to be a robust classifier. The well-known human face database Yale_B is used for verifying system performance, and the recognition rate can achieve to 99.13%. As far as we known, the recognition rate is better than all of the published literatures.