EA-ConvNeXt: An Approach to Script Identification in Natural Scenes Based on Edge Flow and Coordinate Attention

In multilingual scene text understanding, script identification is an important prerequisite step for image recognition. Due to the complex background of images in natural scenes, severe noise, and common symbols or similar layouts different language families, problem has not been solved. This paper proposes a new method based on ConvNext improvement, namely EA-ConvNext. Firstly, generating edge flow map from original proposed, which increases number scripts reduces noise. Then, feature information extracted by convolutional neural network ConvNeXt, coordinate attention module proposed enhance description spatial position vertical direction. The public dataset SIW-13 expanded, Uyghur added, named SIW-14. improved achieved rates 97.3%, 93.5%, 92.4% datasets CVSI-2015, MLe2e, SIW-13, respectively, 92.0% expanded SIW-14, verifying superiority this method.