Structured text detection method and system

Abstract

The invention discloses a structured text detection method and a structured text detection system. The method comprises the following steps: inputting a structured text picture to be detected and a group of text area templates into a trained convolutional neural network; obtaining actual positions of a group of areas to be detected of the structured text picture to be detected by processing of the convolutional neural network, wherein the position of each template in the group of text area templates is a mean value of the positions of corresponding text areas in multiple structured text pictures of the same type of the structured text picture to be detected; the positions of the group of text area templates are determined as the group of areas to be detected of the structured text picture to be detected by the convolutional neural network. According to the method and the system, the detection accuracy is ensured, meanwhile, the calculated amount is reduced as much as possible, and the structured text detection efficiency is greatly improved.

Description

technical field [0001] The present application relates to the field of image processing, in particular to a structured text detection method and system. Background technique [0002] Structured text refers to text with a basically fixed layout structure, such as ID cards, passports, motor vehicle driver's licenses, bills, etc. In the digital age, in order to enter this information into the computer, people often need to manually type and spend a lot of time. In order to save time, people began to use the method of taking pictures of the certificates and using computer vision technology to automatically obtain the text from the pictures. This method is generally divided into three steps: first, intercept all the structured text in the picture as a whole and turn it to the right, so that it fills the whole picture, cut it to remove the background area, and turn it to the right to make the skewed picture straight; second , to detect all regions containing text information; th...

AI Technical Summary

Problems solved by technology

Since the convolutional feature map output by the general deep convolutional neural network has a large area, and each position corresponds to several anchors, the total number of classification scores and adjustments of the anchors that need to be calculated is huge, plus post-processing, making the calculation Too expensive and takes too long

Images