Book Recognition from Color Images of Book Shelves

This paper describes a new approach for book recognition, where color images of the spines of books are analyzed to identify books displayed on book shelves. First, edges of HSY images are unified to one edge image considering the color features of the spines of each book. The color information in the images is effective for segmenting images to robustly extract book boundaries. The approach may resolve the edge extraction problem that occurs in analyzing grayscale images. Then, book title character string regions are extracted using a min/max filter. The filter enables us to extract the regions while avoiding the effects of shading distribution due to non-uniform lighting and the convex surfaces of book spines. In the last step, the cumulative projection profile of each title character string region is compared with that in a database, and the title can be determined without having to recognize each character one by one. The approach presented in this paper makes it possible to visually inspect book shelves, and paves the way for the development of an automated system for managing books in bookstores, libraries, storerooms, and offices.