Deep Learning-Based Image Segmentation System Using SegFormer and U-Net for General and Medical Image Analysis

Authors:

Mrs.Suvvada.B.V.Varalakshmi, Assistant Professor, Information Technology, MVGR College of Engineering.

Gandipilli Ashritha Padmavathi, Information Technology, MVGR College of Engineering,

Kolaparthi Jyothika Syamala Srivalli, Information Technology, MVGR College of Engineering,

Thonangi Thanuja, Information Technology, MVGR College of Engineering,

Tentu Yerram Naidu, Information Technology, MVGR College of Engineering.

Abstract - Image segmentation is a fundamental task in computer vision that involves dividing an image into meaningful regions for better analysis and understanding. This project presents a deep learning-based segmentation system capable of handling both general and medical images. The system utilizes SegFormer, a transformer-based architecture, for segmenting normal images and U-Net, a convolutional neural network, for medical image segmentation. The system is implemented using Python and deployed as a Flask-based web application that allows users to upload images and visualize segmentation results. It processes input images, generates segmentation masks, and produces both segmented and masked outputs. SegFormer effectively handles complex scenes by capturing global contextual information, while U-Net accurately detects fine details in medical images. Experimental results demonstrate that the system achieves high accuracy and flexibility compared to traditional approaches. This work highlights the importance of deep learning in real-world applications such as healthcare diagnostics and computer vision systems.

Keywords: image segmentation, deep learning, segformer, u-net, computer vision, medical imaging