Equation Detection in the Camera Captured Handwritten Document

One of the most important tasks in the realm of document analysis and recognition is the detection of equations in documents that were acquired using a camera. The procedure includes several steps, including pre-processing of the images, segmentation, feature extraction, and classification. The suggested method comprises taking a user-provided input expression image and classifying it into one of three types of equations: simple, complex, and highly complex. By choosing a decision boundary set off from the initial hyperplane, the SVR algorithm encodes the image, producing a model that fits the data better. The result is then obtained by character-wise segmenting the image and comparing it with trained models. Two recurrent neural networks make up the RNN encoder-decoder that is used. One RNN creates a fixed-length vector representation from a sequence of symbols, and a different RNN decodes that representation into a different sequence of symbols. 1900 images containing various equations made up the dataset utilized for training, validating, and testing the SVR and RNN. The accuracy of the system was about 93.64%.