Based on the theoretical basis of the strong correlation between character form and writing style,a handwritten Uy-ghur recognition model based on self-adaptive multi scale style generalization and integration(HTR-MSS)was proposed to im-prove the style adaptive ability.Transformer was improved to extract multi-scale stroke features,the multi-scale style adaptive module was designed to extract the multi-scale sequence features,and the feature generalization fusion module was constructed to deeply integrate the stroke features and sequence features to improve the recognition effect.WER and CER are declined by 3.75％and 0.19％on handwritten Uyghur test datasets compared to that of other base models.It is also validated on IAM pub-lic datasets.
维普
万方数据