Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

A method for handwritten document retrieval based on machine learning

A technology of document retrieval and machine learning, applied in the field of deep learning, to achieve the effect of eliminating human error

Active Publication Date: 2020-09-22
SOUTH CHINA UNIV OF TECH
View PDF2 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] In order to solve the problems existing in the retrieval and matching of existing handwritten documents, the present invention proposes a handwritten document retrieval method based on machine learning, which does not need to perform preprocessing such as word-wise or line-by-line segmentation on the original document image, and directly uses the index text image to Retrieve the corresponding text in the handwritten document, find out the document location where the text to be indexed, and avoid the human error introduced in the segmentation preprocessing

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
  • A method for handwritten document retrieval based on machine learning
  • A method for handwritten document retrieval based on machine learning
  • A method for handwritten document retrieval based on machine learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0027] combine figure 1 , the retrieval method of the present invention has two inputs, one input is a document picture, the other input is an index text picture, and the output is the position of the document where the text is located, and the retrieved text can be marked in the document through post-marking. Searching for a match is a two-step process:

[0028] In the first step, SIFT feature extraction is performed on the two input images. For the indexed text image, we extract key points from the entire image, and use descriptors to represent the features of the indexed text image. SIFT is used to detect and describe local features in pictures. It looks for extreme points in the spatial scale, and extracts its position, scale, and rotation invariance; pictures use Gaussian filters at different scales. Perform convolution, and then use continuous Gaussian blurring image differences to find key points, which are based on the maximum and minimum values ​​of the Difference o...

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 present invention is a handwritten document retrieval method based on machine learning, comprising the steps of: performing SIFT feature extraction on an index text picture, performing key point extraction on the entire picture, and using a descriptor to represent the feature of the picture; performing SIFT on each current page document picture feature extraction, using descriptors to represent key points; matching the extracted descriptors, selecting the text in the candidate frame closest to the descriptor of the index text image, and using the area selected by the candidate frame as a candidate set, and then Use the convolutional neural network to perform further feature matching on the candidate set and the index text image; after retrieving the current page document image, read the next page document image for detection until all pages of the complete document are detected, and the output identifies the index text documentation. The present invention does not need to preprocess the original document picture, and directly uses the index text picture to retrieve the corresponding text in the handwritten document, eliminating the error introduced in the segmentation preprocessing.

Description

technical field [0001] The invention relates to the field of deep learning, in particular to a neural network-based handwritten character matching technology. Background technique [0002] In the study of ancient documents, it is often necessary to scan ancient books into pictures and store them digitally. As the amount of stored data increases, a retrieval mechanism needs to be established. However, since most of the ancient documents are handwritten, the handwriting written by different people is not the same, which puts forward very strict requirements for the matching of the overall document. Moreover, some characters only existed in ancient times and are no longer used in modern times, or the structure and writing of characters have changed with the evolution of society and culture, making it impossible for us to search through computer text input. [0003] Most of the traditional handwritten character recognition often needs to perform line segmentation or word segmen...

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
Patent Type & Authority Patents(China)
IPC IPC(8): G06K9/46G06K9/20G06F16/583
CPCG06F16/583G06V10/22G06V10/462
Inventor 邱梓珩徐向民青春美
Owner SOUTH CHINA UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products