Book Boundary Detection from Bookshelf Image Based on Model Fitting

Several systems based on bookshelf image analysis have been studied for automating bookshelf inspection in libraries and bookstores. In conventional systems, book boundaries are firstly detected for extracting each individual book by using some popular line detection technique, such as Hough transformation. Their detection accuracies, however, are sometimes insufficient. In this paper, we propose a model-based book boundary detection technique for higher detection accuracy. The model is represented as a finite state automaton, each of whose states corresponds to a component of bookshelves, such as boundary and title. With this model, the book boundary detection problem is formulated as a model-based optimization problem where states and local slant angles at all horizontal positions were variables to be optimized. The globally optimal solution is efficiently searched for using a dynamic programming based algorithm. The effectiveness of the proposed technique was shown by several experiments.