以特徵關係為基礎之人臉辨識系統

Abstract

在本篇論文中，我們提出一套人臉辨識的演算法則，主要觀念為利用人臉各特徵之間的比例關係作為辨識的準則。首先由統計關係得到人臉各特徵的大致分佈情形，並利用臉部各特徵具有對稱性的特性，加以搜尋特徵區塊的確實所在位置，然後再針對各個特徵區塊加以擷取特徵，最後我們利用各特徵之間的幾何比例關係作為辨識的依據。經過實驗後得知，當我們把這些比例值乘以一個適當的weighting值後，可以提升辨識的準確度。總而言之，以特徵關係為基礎之人臉辨識系統的優點為可避免許多不必要的搜尋，而直接從區塊內著手，可節省相當多的處理時間，另外，在辨識時也僅需計算各特徵之間的比例，因此運算速度得以提升，同時又能達到相當好的辨識結果。

In this thesis, we have devised an algorithm using a ratio of facial characteristics for human face recognition. First, the distribution of facial characteristics is obtained by statistics and the blocks of facial characteristics are searched by using a property of symmetry. The specific characteristic may be extracted from the blocks of facial characteristics. Finally, these geometric relationships are used to recognize the human face. Furthermore, from experimental comparison, while multiplying a proper weighting value to these ratios, the recognition rate is increased. In a word, the advantages of the statistic characteristics based approach for human face recognition are the avoidance of unnecessary searching and the computation only for the ratio of all the facial characteristics. Therefore, the operation speed of system performance is increased and a high recognition rate may be achieved.