Designing Channel Attention Guided Stacking-Based Contextual Graph CNN For Telugu Text Categorization And Recognition With Character Segmentation Through Deformable DETR++

Abstract

In the present modern life, character recognition is significant to make the handwritten text into machine-readable data. Several research works have been examined and developed for the handwritten character recognition module. However, no research has investigated the regional languages, especially for the Telugu language. Hence, the character recognition on Telugu handwritten text plays an important and challenging task because of its orientation angle, varying writing pattern and different sizes of characters. The accurate identification is still become the complex task for the compound character. Recent studies developed a neural network to evaluate the discriminative qualities from the huge volume of data, which tends to be an effective solution for handwriting recognition. However, there are still some limitations in achieving an accurate recognition outcome, especially for the Telugu character. Hence, in this work, a novel deep learning-based handwritten Telugu character classification and recognition model is designed with segmentation procedures. In the beginning, the significant Telugu text images for the validation are sourced from the benchmark resources. Next, the collected Telugu text images are forwarded to the character segmentation phase. Here, Deformable DEtection TRansformer++ (DDERT++) is employed to carry out the segmentation procedures. Once the images are segmented, and then forwarded to the Telugu text categorization and recognition phase. In this phase, an Adaptive and Channel attention guided Stacking based Contextual Graph Convolutional Neural Network (ACS-CGCNN) is used to carry out the Telugu text categorization and recognition procedure. Moreover, the parameters of ACS-CGCNN are optimized using Mutated Hunting-based Groupers and Moray Eels Optimization (MHGMEO), which used to improve the accuracy of Telugu text categorization and recognition. Later, different investigation are conducted in the developed Telugu text categorization and model over the existing schemes under different experimental conditions.