DKNet: Deep Kuzushiji Characters Recognition Network

Kuzushiji, a cursive writing style, had been extensively utilized in Japan for over thousand years starting from the 8th century. In 1900, Kuzushiji was not included regular school curricula due to change Japanese system. Nowadays natives are unable read historical books that were written using language. Therefore, libraries and museums have decided build digital copies of documents Due limited number trained experts, researchers machine deep learning models convert into modern script can be easily by human beings. However, existing techniques suffer over-fitting gradient vanishing problems. To overcome these problems, an efficient characters recognition network (DKNet) is proposed. Initially, remove noise training images, trilateral joint filter applied. Contrast adaptive histogram equalization (CLAHE) then applied enhance visibility filtered images. Thereafter, pre-trained MobileNet extract features characters. MobileNet’s final layers removed, including fully connected layer softmax. The flatten input. A classification used with Rectified linear units (ReLUs) dropouts. Dropouts generalize model, thus preventing problem. Finally, softmax activation function employed provide results. test proposed actual first segmented Maximally stable extremal regions (MSERs) convexhull-based segmentation approach. Segmented recognized DKNet. Extensive comparative analyses reveal DKNet achieves better performance than competitive terms various metrics. An Application Programming Interface (API) also designed ancient heritage character help end-users.