Scene text recognizing method and device

Abstract

The invention is applicable to the technical field of text identification and provides a scene text recognizing method and device. The method comprises obtaining candidate text components in a text image, preprocessing the obtained candidate text components and extracting component regional context features; obtaining the global features in the text image; inputting the component regional context features and the global features into a pre-established credit degree network model; according to a character recognizer in the pre-established credit degree network model, recognizing characters that the component regional context features belong to; pairing the recognized characters to compose text lines or entries. The scene text identifying method avoids the situations that scanning a large number of sliding windows line by line leads to huge operations as well as sliding window scanning is restricted to horizontal text detection, thereby reducing operations; by achieving detection in multiple directions and through the credit degree network, the error detection rate can be effectively reduced, and the text recognizing efficiency and accuracy can be improved.

Description

technical field [0001] The invention belongs to the technical field of character recognition, and in particular relates to a method and device for recognizing scene characters. Background technique [0002] Scene characters are characters that exist in various scene images in daily life. Usually, in a scene image where text and other objects coexist, users tend to pay more attention to the text in the scene image first, so extracting the text that appears in the scene image and recognizing the text in the scene image will provide users with meaningful s help. [0003] Generally, the text detection system that extracts the text appearing in the scene image and recognizes the text in the scene image can be divided into two categories. [0004] The first type is a text detection system based on the sliding window model. The basic principle of this model is to use sliding windows of different sizes and shapes to traverse different regions of the scanned scene picture. [0005...

AI Technical Summary

Problems solved by technology

[0014] The purpose of the embodiments of the present invention is to provide a method for recognizing text in a scene, aiming to solve the problem of large amount of calculation, single scanning direction, easy generation of many incorrect candidate components and false detection of many similar texts in the existing text detection system ingredient question

Images