使用橫向分割的方法進行人臉檢測

Abstract

人臉檢測的方法相當多樣化，不論是使用向量的方式來計算或是用類神經網路的方法來訓練，都會有非常大的計算量，而在一些需要 "即時" 的應用上，例如門禁系統的身份辨認，如果要花太久的時間來等待，使用者一定會認為這個系統不甚方便。 本論文試圖以較為直覺的方法來進行人臉檢測的工作。以邊緣檢測為基礎，將圖片予以橫向分割成帶狀長條，便於找出具有邊緣性質的區域範圍，其次使用樣版比對的方式來找出臉的重要特徵--眼睛--的可能位置，然後再根據人臉幾何結構原則找出最可能的具有配對關係的眼睛位置，最後以 Hough 轉換來求得眼睛虹膜的中心。而嘴巴的定位則以眼睛為基準，在符合人臉的幾何結構原則下，來尋找嘴角的定位與嘴唇的外緣。

There are several methods for face detection. No matter using eigenface-based method or neuro network-based training, it usually spend a lot of time on calculating. In some applications need "real time" reaction, if someone have to wait for a long time, he will think that the application is not convenient. We propose a simple method for face detection. The method is based on edge detection followed by dividing the result map after edge detecting into several horizontal strips. In each strip, we can find some blocks with obvious edges. We call the blocks with interested blocks. Then, using iris template to compare everywhere in interested blocks to find iris candidates. With face geometrical model, we can find some pairs of the iris candidates. Then, applying Hough transform on the pairs to find the centers of iris. With the centers of iris, we have a baseline to decide the range of mouth. After that, we can find the location of mouth and its outline.