Using a Hybrid Model for Image Recognition of Handwritten Text

Abstract

In the recognition of Uzbek handwritten letters, an important issue is segmentation and its recognition. The curvature of the handwriting, cursive and word-to-word problems, incorrect placement of diacritics, and the presence of ups and downs lead to the processing of the segmentation problem. In the manuscripts, the performance of linear and word segmentation is shown based on results obtained from software based on the CNN+LSTM+CTC neural network model.