Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Spine text recognition method and device based on deep learning and storage medium

A text recognition and deep learning technology, applied in the field of neural network text recognition, can solve the problems of inconsistent fonts, spine leakage recognition, blurring, etc., to achieve high-precision recognition, ensure accuracy, and achieve the effect of automation

Pending Publication Date: 2020-09-18
GUANGDONG POLYTECHNIC NORMAL UNIV +1
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] CRNN model: This method is a general model that can realize the recognition of text in ordinary scenes. The accuracy of text recognition is higher for text with simple background and single layout rules, but the recognition performance for text on the spine is poor, because the spine has blurred text and fonts. Due to the characteristics of variety and uncertain typesetting rules, this method may miss or misidentify spines for blurred, non-uniform fonts, and complex typesetting.
This makes this method unable to perform high-precision recognition in the spine recognition scene

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Spine text recognition method and device based on deep learning and storage medium
  • Spine text recognition method and device based on deep learning and storage medium
  • Spine text recognition method and device based on deep learning and storage medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0026] This implementation further discloses a spine text recognition method based on deep learning, which specifically includes the following steps:

[0027] Step (1), making a synthetic spine training set, a single-word multi-font training set;

[0028] The process includes collection of bibliography, publishing house, author text information, data cleaning, synthesis of spine text images in various font forms and backgrounds; collecting 2536 commonly used characters in Chinese and English numbers, and making a variety of texts including artistic fonts, street view fonts, etc. Font word image dataset.

[0029] Step (2), collect real spine pictures in the library, use the CTPN model to carry out text detection and cutting, labeling to obtain the real spine data set for a single spine;

[0030] Step (3), build the network model of ResNet10+BiLSTM+CTC;

[0031] The role of ResNet10 in this model is mainly to extract the spine features. In this paper, the improved ResNet netw...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a spine text recognition method based on deep learning. The spine text recognition method comprises the steps: manufacturing a synthesized spine, single-character multi-font and a real spine data set; performing text detection on the single book spine based on a CTPN model; establishing a spine text recognition model STRNet based on ResNet10 according to the text fuzzy characteristics; manufacturing an SK-NNS classifier according to various font characteristics; training an STRNet classifier and an SK-NNS classifier on the basis of the data set made by the user; integrating the STRNet model and the SK-NNS model on the basis of the false identification probability point wp, and performing SK-NNS secondary identification on a single word which may be wrongly identified by the STRNet on the basis of the log _ softmax prediction probability; after text recognition post-processing, performing entry grammar error correction by using a natural language processing technology, so as to ensure the recognition accuracy. The STRNet can extract more spine text features, the SK-NNS can recognize single characters of different fonts, the advantages of the two models are complementary, and the application accuracy of the method in a spine text recognition scene is higher.

Description

technical field [0001] The present invention relates to the field of neural network text recognition, in particular to a spine text recognition method, device and storage medium based on deep learning. Background technique [0002] The digitalization of books is the trend of future development, but digital books cannot replace the touch quality brought by paper books. At present, paper books are still the main component of book resources. However, the routine inventory work of millions of books requires a lot of manpower and material resources. Using computer vision technology to simulate human recognition capabilities is a development direction for building an intelligent library. High-precision text recognition on book spines is an important sub-project. . The traditional way of recognizing text on the spine by manual or barcode is low in efficiency, high in cost and high in error rate. In recent years, text recognition technology based on deep learning has been used in d...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06K9/32G06K9/34G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06V20/63G06V30/153G06N3/047G06N3/045G06F18/24147G06F18/2415Y02D10/00
Inventor 蔡君付鸿添廖丽平陈庆珊
Owner GUANGDONG POLYTECHNIC NORMAL UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Eureka Blog
Learn More
PatSnap group products