Automatic car make recognition in low-quality images

Abstract

This paper presents a robust system for automatic car make recognition in real-traffic images of car front, featuring low contrast and compression-based distortions. The system is designed to distinguish and classify a variety of car makes by means of Scale Invariant Feature Transform pattern recognition and matching over a reference database of car brand images. The system framework consists of image preprocessing techniques yielding a car brand region, feature extraction and description, pattern matching procedure and multicriteria decision-making process. The knowledge database is opened and easy to extend in order to cover an increasing number of car makes. Described approach stands for a part of an expert system for car type, make and color recognition, to be designed and build for real traffic supervision.