書棚画像からの書籍の背文字領域の抽出について

概要

論文の詳細を見る
In this paper, we examined the extraction of character regions from bookshelf images to assist in making a book database system. The extraction algorithm has two processes, 1) an extraction of the book images from the bookshelf image, 2) an extraction of character regions from the book images. In the first stage, the bookshelf image is divided into several sectors based on the direction of density gradient of the image pixels. Then, by applying the Hough translation, the straight lines composing the edge lines of the book images can be detected. In the second process, we use the peripheral distribution of the book image to detect the width and height of each character. Using 23 bookshelf images containing 124 books with 1227 characters, we obtained an extraction rate of book images of 93.5 (%) and an extraction rate of character regions of 89.9 (%).
岡山理科大学の論文