Book identification method and system based on spine visual information

Abstract

The invention provides a book identification method and system based on spine visual information, and the method comprises the steps of collecting a book spine image on a library shelf, and carrying out the manual marking of the book spine image, so as to construct a spine segmentation and spine classification data set; constructing a convolutional neural network for spine picture depth feature extraction, and performing training by using the training data to obtain a feature extraction model; shooting a picture on one side of a spine of a book on a shelf, performing instance segmentation on the spine, acquiring a visual feature vector corresponding to the spine picture by using a trained model, and matching the visual feature vector with a library collection database to identify the category of the book corresponding to the spine. According to the invention, the spine picture of the book is recognized based on the deep learning algorithm, all visual information of a spine target is utilized, the method is not limited by a dictionary set on which a character recognition method depends, library collection of newly added books is supported, and higher accuracy and better robustness and expandability are achieved; the spine pictures of a series of books on the shelf can be identified in batches.

Description

technical field [0001] The invention relates to the field of book information management, and in particular to a book recognition method and system based on spine visual information. Background technique [0002] At present, the management of book information still stays in manual identification based on human eyes, manual sorting of book information, or based on radio frequency identification technology, customized electronic tags (RFID) and non-contact signal receivers to sort out book information, or based on text recognition technology, pictures Acquisition equipment and optical character recognition algorithms, character recognition algorithms based on deep learning, sort out book information. [0003] Among them, the recognition based on human eyes is to manually pick, place, identify and classify the books on the shelf. Based on radio frequency identification technology: install electronic tags (RFID) for each book, enter book information, and use non-contact equipme...

AI Technical Summary

Problems solved by technology

For example, the book sorting system and the book collection station (CN201610632579.2), the library book picking robot (CN104552230A) and the automatic collection method and system of publication digital resources (CN104424271B), although they are based on optical character recognition, they do not involve The identification of the spine can only classify the book by identifying the cover of the book or the content of the book

However, in actual use, books are often densely placed on bookshelves, and only the spine is exposed. It is difficult to identify the book category of pictures containing multiple spines in the existing technology.

[0004] The current technology still has the following technical defects. For example, recognition based on human eyes will consume huge time and labor costs, which greatly limits the improvement of work efficiency.

Based on radio frequency identification technology requires heavy pre-construction work, relies on proprietary equipment and systems, and is expensive

Character-based recognition technology is sensitive to the thickness of the book, the degree of wear and the variety of artistic characters, and has poor stability; it is essentially a classification of characters, which is heavily dependent on the scope of the dictionary set, that is, it cannot recognize unfamiliar languages ​​and fonts that are not in the dictionary set. not scalable

Images