Feature Identification based Automatic Sign and Photo Detection in a Document

This naner proposes a method for automatically detecting images and signatures in uploaded documents. The prime objective is to assist the organization in reducing the number of faulty admit cards issued to trainees who have limited computer knowledge by providing an automated software service that can rearrange the unmatched uploads made by trainees on the portal during their exam registration process. This will be helpful not only for trainees but also for other applications that require the unloading of documents containing a signature and a photograph. In addition to a trained neural network for face and signature classification, Haar cascades are utilized to improve facial feature identification. For signature detection, three distinct edge detection models are utilized to improve signature feature identification. The face detection model uses two datasets. namely LFW and FER-2013, while the signature detection model uses three datasets of signatures in three different Ianguages. The accuracy of the face detection model is 99.89 percent, while the accuracy of the signature analysis model is 98.56 percent. Finally, a weighted average of both models is calculated to produce an output with improved validity.