donproc / book-identifier

A Matlab app for identifying books from bookshelf images.

Book-Identifier

This is a Matlab app for identifying books from images of bookshelves. It was created as part of a university project for a computer vision course.

The repository includes the code, a Matlab application for testing the code, a webpage describing the project, a powerpoint presentation, and a pdf report describing the project in detail. For further details refer to the report/webpage/presentation.

The abstract:

While many websites provide review-based book rankings, brick and mortar stores do not provide easy access to such data. In this project, we develop a system for identifying books from a photo of a bookshelf. This information can then be used to extract any required book-related data from the internet, thus providing the clients of brick-and-mortar bookstores with the same kind of information as is available online. Our system extracts book spines from the photo, segments the text from the background, and groups individual characters into words. An implementation is also provided for training a classifier for character recognition. For segmentation, we propose and implement a simple method suitable for most book spines. For the grouping of characters into words, we develop a new algorithm, suitable for text with a variety of orientations and sizes. Our system achieves successful spine separation in around 94% of cases, successful segmentation of text in 77% of the spines, and correct word grouping for 74% percent of successfully segmented words.