Vietnamese scene text detection method and device fusing edge information and text enhancement

Abstract

This invention discloses a method and apparatus for text detection in Vietnamese scenes that integrates edge information and text enhancement. The method includes the following steps: S01. Inputting the image to be tested into a backbone network to extract multi-layer features, and extracting edge detail information by an EIEM module based on a channel attention mechanism, and fusing the text edge detail information with the first-layer features; S02. Inputting the top-layer features extracted from the backbone network into a TREM module to extract global context information and character dependencies, adjusting the features extracted from each layer of the backbone network according to the features output by the TREM module, and fusing the feature maps of different layers to form a text region enhanced feature map; S03. Performing text post-processing to obtain a probability map and an adaptive threshold map, performing variable binarization to obtain an approximate binary map to determine the boundaries of the text box. This invention has the advantages of simple implementation, high detection accuracy, and strong robustness.

Images