Text detection method based on channel grouping attention mechanism

Abstract

The invention provides a text detection method based on a channel grouping attention mechanism, and aims to solve the problem that targets of different scales adopt preset boxes of different scales ona characteristic spectrum, and a higher-resolution prediction characteristic spectrum with more spatial information before fusion cannot well represent text characteristics and is poorer in effect when being directly input into a detection head. After experiments, the applicant finds that an attention module is introduced into a prediction characteristic spectrum with a quarter resolution, the text offset prediction and candidate box regression effects of the TextBoxes_plusplus algorithm under the quarter scale are improved, and therefore the method adapts to the changeable length-width ratioof the text, missing detection and false detection are reduced, and the characteristic robustness is well enhanced. Spatial information is reserved, and small target text detection precision is improved. In addition, convolution and grouping convolution are used for replacing full connection in encoding and decoding of the attention mechanism, the parameter quantity and the calculation complexityare reduced, and compared with attention of a common channel, the attention mechanism is efficiently achieved under the condition of performance approximation.

Description

technical field [0001] The invention relates to text detection technology, in particular to channel attention in text detection. Background technique [0002] OCR (Optical Character Recignition) optical character recognition, with the development of digital multimedia technology, now generally refers to image text recognition, and text information needs to be extracted in various visual data analysis and applications. Text detection aims to find areas where text is located. The purpose of character recognition is to identify the character category of the region of interest. Reading and recognizing text on computer pictures, inputting data such as periodical collection forms into computers for processing and storage, all support the rapid development of text detection and recognition methods. Further combined with emerging directions such as advertising recommendation, material classification, and video review, OCR is full of enduring vitality. The current mainstream OCR t...

AI Technical Summary

Problems solved by technology

This type of method treats the text as an object and has good performance, but the aspect ratio of the text itself changes drastically, and the detection accuracy of the text may have problems such as skew and distortion.

Images