Natural scene character detection method based on attention mechanism

Abstract

The invention discloses a natural scene character detection method based on an attention mechanism, and the method comprises the steps: designing a convolutional neural network model for extracting a text target according to the feature information of a text center block and a stroke region, and training the model through employing the text center block and stroke information as supervision data; in a model test stage, respectively inputting a test image into the text center block model and the stroke model to obtain a probability graph of a text center block and a word stroke area; a final text area is obtained through reasoning and marked, and a scene image text detection task is completed; according to the method, the problems that in the prior art, in scene character detection, direction information of a curved text is not prone to regression, adhesion between adjacent different text lines and information redundancy generated by multi-level feature integration occur are solved.

Description

technical field [0001] The invention relates to the field of computer analysis, in particular to a method for detecting characters in natural scenes based on an attention mechanism. Background technique [0002] As the carrier of human knowledge and information, text widely exists in real daily life scenes. Extracting the text information embedded in the scene image is very valuable and beneficial in many applications based on image content information. The text extraction technology in the scene image is used in blind navigation, blind reading, image retrieval and labeling, human-computer interaction, It has broad application prospects in unmanned driving and other scenarios. Scene text detection is to determine the specific position of the text in the image, and text recognition is to recognize the text information in the bounding box into scale text. Scene text detection plays an important role in extracting and understanding text information in scene images, and its per...

AI Technical Summary

Problems solved by technology

[0005] The direction of curved text is not easy to regress: the scene text detection method based on box regression needs to regress the direction information when solving the direction of multi-directional scene text examples; however, for text examples of arbitrary shapes, such as curved shapes, the direction information cannot be make regression;

[0006] Glue between different text lines: The scene text detection method based on semantic segmentation has achieved good performance in solving the scene text problem of any shape and any direction; however, if different text lines are relatively close to each other, it is easy to Causes glue between lines of text;

[0007] Multi-level feature integration produces information redundancy: the semantic segmentation-based scene text detection method uses shallow and deep multi-level features when predicting text region information; however, text objects are mainly concentrated in deep features; in addition, feature integration In the process, the importance of different features is not considered,

Images