應用差值影像於人臉辨識之研究

Abstract

本論文提出一個名為 "差值影像前處理" 的方法，目的在改善一般人臉辨 識系統在辨別受到整體性緩慢灰階變化影響的人臉影像時的效能，例如 : 光線條件的變化 ......等等。從差值影像的特性而言，臉部特徵所造成 的灰階差異會比緩慢灰階變化所造成的效果明顯，因此應用此種差值影像 來做辨識可以有效地解決影像退化以及光線變化所造成的影響。除此之外 ，由於差值影像中的灰階動態範圍變小，我們可以利用這種特性來加速辨 識程序以及達到資料壓縮的目的。在本論文中，我們把差值影像前處理應 用到兩種不同的辨識系統 - Eigenface method 和 Fisherface method， 用來驗證其效能與可行性。實驗結果顯示這兩種辨識系統無論在辨識一般 、退化或是有光線變化的影像時，仍具有很高的辨識率。 A preprocessing, differential image generation, is proposed in this thesis to improve the performance of face recognition systems for recognizing face images degraded by slow-changing global variation in gray values, such as lighting conditions. The motivation of using the differential image for face recognition is resulted from the observation that main features of a face usually cause more significant changes in the gray values for image pixels in a small neighborhood than that due to slow-changing global variation in gray values; therefore, using the differential image may alleviate the above image problem. Using the differential image may also speed up the recognition progress and save the data storage, since the dynamic range of the gray values of the differential image is much smaller than the one of the original image. In this thesis, the proposed method are applied as preprocessing for two face recognition systems, Eigenface and Fisherface methods. Experimental results show that the proposed method has good recognition rates when recognizing degraded images as well as normal ones.