Multi-lingual Offline Signature Recognition Based on LOMO Feature
Aliya Rexit, Mahpirat Muhammat, Ruxianguli Abudurexiti, Yusnur Muhtar, Kurban Ubul · 2022
Despite the development of electronic signatures, most people still prefer to use handwritten signatures. Signature recognition is used in banks, judiciary, financial transactions, and many other fields. Offline signature recognition is well established for common languages like English but is still in the developmental stage for minority languages like Uyghur. This paper proposed a multilingual hybrid handwritten signature recognition method to address this problem based on local maximum occurrence features. Firstly, offline handwritten signature datasets in Uyghur and Chinese were established. Next, the signature images were pre-processed by grayscale, normalization, bilateral filtering, and binarization. Then, we extracted the Local Maximal Occurrence Representation (LOMO) features and reduce the extracted features' dimensions was used principal component analysis(PCA). Finally, in the stage of classification, we utilized the random forest and K-nearest neighbor algorithm. The experimental results show that the best recognition accuracy of this method was 98.4% for the self-built signature dataset, which has a high application value.