DeblurGAN-CNN: Effective Image Denoising and Recognition for Noisy Handwritten Characters

Many problems can reduce handwritten character recognition performance, such as image degradation, light conditions, low-resolution images, and even the quality of capture devices. However, in this research, we have focused on noise images that could decrease accuracy recognition. types penalties influence for example, low resolution, Gaussian noise, contrast, blur. First, research proposes a method learns from noisy synthesizes clean using robust deblur generative adversarial network (DeblurGAN). Second, combine DeblurGAN architecture with convolutional neural (CNN), called DeblurGAN-CNN. Subsequently, two state-of-the-art CNN architectures are combined DeblurGAN, namely DeblurGAN-DenseNet121 DeblurGAN-MobileNetV2, to address many enhance performance images. Finally, DeblurGAN-CNN transform characters new recognize simultaneously. We evaluated compared experimental results proposed existing methods four datasets: n-THI-C68, n-MNIST, THI-C68, THCC-67. For n-THI-C68 dataset, achieved above 98% outperformed other methods. an 97.59% when AWGN+Contrast was applied digits. THCC-67 dataset. The result showed 80.68%, which is significantly higher than method, approximately 10%.