A Method for Extraction and Recognition of Isolated License Plate Characters

TL;DR

This paper introduces a novel method for extracting and recognizing isolated license plate characters using Difference-of-Gaussian for detection and a new feature vector called AGPV for recognition, demonstrating robustness to various image conditions.

Contribution

The paper presents a new approach combining DOG-based detection with AGPV features for improved license plate character recognition.

Findings

AGPV features are invariant to scale and rotation.

The method is robust to noise and illumination changes.

Connected component analysis effectively extracts candidate characters.

Abstract

A method to extract and recognize isolated characters in license plates is proposed. In extraction stage, the proposed method detects isolated characters by using Difference-of-Gaussian (DOG) function, The DOG function, similar to Laplacian of Gaussian function, was proven to produce the most stable image features compared to a range of other possible image functions. The candidate characters are extracted by doing connected component analysis on different scale DOG images. In recognition stage, a novel feature vector named accumulated gradient projection vector (AGPV) is used to compare the candidate character with the standard ones. The AGPV is calculated by first projecting pixels of similar gradient orientations onto specific axes, and then accumulates the projected gradient magnitudes by each axis. In the experiments, the AGPVs are proven to be invariant from image scaling and rotation, and robust to noise and illumination change.