Segmentation Based Method For Handwritten Connected Digits Using Image Encryption Technique

Abstract

This project falls within the field of deep learning, a branch of artificial intelligence that focuses on developing computer models that learn and evolve autonomously from data. The aim of this project is to improve the process of number recognition using deep learning techniques. The benefits of number recognition through deep learning are significant and diverse. For example, it can be used in applications such as handwriting recognition, number recognition in images, understanding digital information in images and documents, and even enhancing security and identity verification through devices recognizing secret numbers or numeric codes. In this project, a Convolutional Neural Network (CNN) model was employed, which is a type of artificial neural network specialized in image processing and recognition. The model was trained using the CVL dataset, which contains images of individual numbers. Subsequently, an image segmentation technique was used, where the number sequence was divided into individual parts for each number using the trained model. Finally, the predicted numbers were collected to obtain the complete number sequence. Multiple methods were proposed to improve the segmentation process, including sliding window and edge detection in the image. The two proposed methods were combined, and the results were monitored to enhance prediction quality. To further enhance the segmentation quality, an image encoding technique was used. Techniques such as filters and image enhancement were employed to improve the image quality before segmentation.