- 期刊
影像資料庫在人臉辨識上的應用－前科犯資料庫
An Application of Image Database on Facial Recognition-The Database of Lawbreakers
摘要
本研究旨在建構一前科犯的人臉影像資料庫，警方在取得目擊者所描述之人臉影像後，即可利用此資料庫的人臉辨識搜尋系統找出可能的嫌疑犯。首先我們建立前科犯的影像資料，將以警方為人犯拍攝的標準檔案照片，直接掃描成影像檔，再取出臉部各器官，包括雙眼，雙眉，唇形，臉型輪廓及側臉輪廓等，經影像基本處理，邊緣強化，索貝爾邊緣檢測，哈克轉換及細線化等處理，取出各器官影像特徵邊線。再以傅立葉轉換求出傅立葉描述子，經正規化處理，取得正規化特徵頻譜，再取其第二到第五分量，以建立該器官之資料庫。其次，為各關聯建立空間索引。目擊者畫像亦以相同技術處理，再以之查詢空間索引，俾縮小偵察範圍，各器官之重要性加權亦一併於文中探討。
並列摘要
The goal of this project is to construct a human image database of lawbreakers. After the police get the human image drawn by the witness, they can use the human image database and retrieval system to find out the possible upsets. At first, we should construct the image data of lawbreakers, then scan the standard mug shot taken by the police and transfer it to the image file. After retrieving the features of face organs including eyes, eyebrows, lips, outline of face and profile, by using edge enhancement, Sober edge detection, Hough transform and thinning techniques, we can obtain the outlines of each organ. The outlines are then transform into Fourier descriptors with Fourier Transform. After that a normalized feature spectrum is generated. The second to fifth spectrums are extracted to construct organ relations. We also construct the spatial index for every relation. The human image drawn by the witness is also proceeded by the same techniques described above. We use the image to query the spatial index to downsize the searching space. The importance of weight of each organ is also discussed in this article.