%0 Journal Article %T An Effective Color Addition to Feature Detection and Description for Book Spine Image Matching %A Spencer G. Fowers %A Dah-Jye Lee %J ISRN Machine Vision %D 2012 %R 10.5402/2012/945973 %X The important task of library book inventory, or shelf-reading, requires humans to remove each book from a library shelf, open the front cover, scan a barcode, and reshelve the book. It is a labor-intensive and often error-prone process. Technologies such as 2D barcode scanning or radio frequency identification (RFID) tags have recently been proposed to improve this process. They both incur significant upfront costs and require a large investment of time to fit books with special tags before the system can be productive. A vision-based automation system is proposed to improve this process without those prohibitively high upfront costs. This low-cost shelf-reading system uses a hand-held imaging device such as a smartphone to capture book spine images and a server that processes feature descriptors in these images for book identification. Existing color feature descriptors for feature matching typically use grayscale feature detectors, which omit important color edges. Also, photometric-invariant color feature descriptors require unnecessary computations to provide color descriptor information. This paper presents the development of a simple color enhancement feature descriptor called Color Difference-of-Gaussians SIFT (CDSIFT). CDSIFT is well suited for library inventory process automation, and this paper introduces such a system for this unique application. 1. Introduction Taking inventory is a daunting task in any industry, especially when the number of items reaches into the multimillions, as is the case with most major libraries. It turns into a very challenging and costly task because each item has to be accounted for without the benefit of automation. A comparison done by the On-line Computer Library Center shows that libraries in the United States alone circulate more books every day than the shipping giant FedEx delivers packages. Approximately 5.4 million books are checked out daily from libraries across the US. Furthermore, libraries worldwide hold an estimated 16 billion volumes, and this number continues to grow. Even allocating just one second per book, a full inventory would require over 507 man-years. When equipment such as a barcode scanner is used, each book must be taken off the shelf, its cover opened, the barcode scanned, and then reshelved. Even with such improved technology, the amount of time and labor required is still substantial. Another promising alternative is to use radiofrequency identification (RFID) chips. This approach, however, requires replacing existing call numbers, special labels, or barcodes, constituting a %U http://www.hindawi.com/journals/isrn.machine.vision/2012/945973/