Enhanced Sw Algorithm for Text Extraction and Segmentation in Indian Vehicle License Plate

Abstract

Detecting and recognizing characters on vehicles' license plates has always posed a significant challenge due to various conditions. These conditions include different illumination levels, unavoidable data acquisition noises, and demanding scenarios, such as the need for real-time performance in advanced Intelligent Transportation Systems (ITS) applications. This research paper presents a method for the application of Vehicle License Plate Character Segmentation and Recognition (VLPCSR) that achieves high accuracy and real-time performance. The proposed system is specifically designed for Indian vehicle license plates, which exhibit characteristics such as varying resolutions and layouts, limited digits/characters, diverse background colors, and different font sizes. To address these challenges, the system utilizes the Sliding Window Algorithm (SWA) based on Alexnet, which automatically extracts Persian characters from input images in two steps. A diverse range of vehicle images from various conditions, along with a dataset from Kaggle, have been utilized for training and testing. The practical application of the system has been installed in real-life scenarios. Experimental results demonstrate an impressive end-to-end accuracy of 96.33% on a dataset of 300 images. The test data includes both color and grayscale images capturing vehicles at different distances and shooting angles, with variations in brightness and resolution. Furthermore, the system exhibits excellent performance in terms of recall, precision, and the Harmonic mean, making it well-suited for real-world data.