基於樸素貝氏分類器採用FPGA之即時與低記憶體之多人臉偵測系統設計

Abstract

近年來人臉偵測被廣泛應用於各領域中，如人臉識別，圖像聚焦，和監視系統等。本論文提出一個以樸素貝氏分類器為基礎採用FPGA的即時人臉偵測系統，分為特徵擷取、候選人臉偵測、誤判消除三個部分。首先以影像金字塔(Image Pyramid)的方式同步縮小並對每個圖片擷取局部二值圖像特徵;接著局部二值圖像特徵經過樸素貝氏分類器找出候選人臉，最後，利用膚色過濾與重疊人臉消除，去除誤判。偵測結果以VGA形式輸出至螢幕上。 本論文實現人臉偵測系統於FPGA上，由於FPGA平行化處理的特性，在640480解析度下，一張影像的人臉偵測僅需16.7毫秒。且使改良過的局部二值圖像特徵，相對於哈爾特徵(Haar-like features)，可以節省約140倍的記憶體使用量，論文所提及之即時人臉偵測系統確實具有成效，可達九成五以上之辨識率。

In recent years, face detection is widely used in various fields, such as face recognition, image focusing, and surveillance systems. This thesis proposes a real-time face detection system based on naive Bayesian classifier using FPGA. The system divided into three main parts, feature extraction, candidates face detection, and false elimination. First downscale the image to the image pyramid and extract local binary image features from each downscaling image; then features go through the naive Bayesian classifier to identify candidate faces. Finally, use skin color filter and face overlapping elimination to remove false positives. Detection results output to the monitor in VGA. In this thesis, face detection system to implement in FPGA. As a result of the FPGA parallel processing, in 640480 resolutions, the face detection of an image executes within 16.7 milliseconds. And the improved local binary features, compared to Haar features, save around 140 times the amount of memory. The experimental results show that the accuracy rate is higher than 95% in face detection, which implies the proposed real-time detection system is indeed effective and efficient.