Stamp character detection algorithm based on improved TextSnake
Stamp characters are easy to be wrongly detected and missed due to fuzzy characters,variable as-pect ratio,overlapping with background text and other factors.In order to solve these problems,a stamp char-acter detection algorithm based on TextSnake is proposed,which is referred to as the TextSnake-CR algo-rithm.Firstly,a hand-designed receptive field enhancement module is embedded in the feature fusion net-work,so that the shallow features have a larger receptive field of view,and then effectively reduce the false detection rate.Secondly,the color feature extraction module is proposed to extract the color feature of the stamp character,and enhance the detection accuracy of the fuzzy stamp character.Finally,the classification loss function of the model was refined to suppress the interference of background noise on the model,and fur-ther improve the detection performance of the model.The experimental results show that the TextSnake-CR achieves the F score of 90.71%and 81.79%on the public and the homemade stamp datasets,respectively.Compared with other algorithms,the TextSnake-CR effectively improves the detection performance of stamp character.
stamp character detectionTextSnakecolor feature extractionreceptive field enhancementclassi-fication loss function
NETL
NSTL
万方数据
