2D Self-organized ONN model for Handwritten Text Recognition

Deep Convolutional Neural Networks (CNNs) have recently reached state-of-the-art Handwritten Text Recognition (HTR) performance. However, recent research has shown that typical CNNs’ learning performance is limited since they are homogeneous networks with a simple (linear) neuron model. With their heterogeneous network structure incorporating non-linear neurons, Operational (ONNs) been proposed to address this drawback. Self-ONNs self-organized variations of ONNs the generative model can generate any function using Taylor approximation. In study, in order improve level HTR, 2D Self-organized (Self-ONNs) core novel proposed. Moreover, deformable convolutions, which demonstrated tackle writing styles better, utilized study. The results over IAM English dataset and HADARA80P Arabic show operational layers significantly improves Character Error Rate (CER) Word (WER). Compared its counterpart CNNs, reduce CER WER by 1.2% 3.4 % 0.199% 1.244% dataset. benchmark demonstrate outperforms deep CNN models significant margin while use convolutions demonstrates exceptional results. • We propose for Recognition. utilize Self-ONNs. outperform models. shows margin.