An approach to recognize book title from multi-cell bookshelf images

Abstract

There are many conventional methods for book detection and title recognition from bookshelf images. But most of these methods are worked on single row bookshelf images. Here this paper presents a technique for segmenting books spine and recognizing book title from multi-row bookshelf images. Horizontal edges are detected and extracted from the images as to indicate individual rows. These separated row images are used in the next module where vertical lines are extracted in order to segment the book regions. Later all book spine images are converted into binary images. At the next step, small and unwanted objects are removed using region properties and subsequently and extracts titles from individual book spines. Then the characters of the title are segmented and extracted by using bounding box and connected component region. Separated character images are matched or unmatched with the data set images by applying template matching. As a result, the developed new method recognizes the title. The system design as a whole makes a contribution, but the extraction of the book titles from the multi-cell images makes the main principal of this paper. To test the proposed framework various bookshelf images with a variety of conditions are used and results are presented to prove its effectiveness.