Vehicle-Logo Recognition Method Based On Discriminative K-SVD

With the development of modern intelligent city, vehicle plays an important role in modern life.As an important part of the automatic vehicle identification system and and the smart residential area system, intelligent transport systems have caused widespread concern in domestic and foreign researcher,and license plate recognition and vehicle logo recognition is an important component.This article focuses on the study car logo recognition, including car standard location and recognition. Due to the diversity of the subject vehicle and small target areas, vehicle logo positioning is still a technical difficulty, there is no mature applications, so this research has important scientific and practical value.Sparse representation of the signal has the most concise representation capability,this article, a vehicle logo recognition algorithm based on sparse representation has proposed by depth research on the sparse representation theory. The algorithm includes two modules:vehicle logo location and vehicle logo identification,mainly for small and medium domestic road vehicles,studying the key technical of automatic vehicle logo identification system.While kept the speed of vehicle-logo recognition,we improve the reliability of the vehicle-logo recognition, and finally provide effective solutions. Main content of this article is as follows:(1) Pre-processing the image to locate the plate and eliminate the interference of background. Preprocessing includes:gray image, binarization processing, image morphology and edge detection techniques.(2) To realize vehicle plate location and vehicle logo location,first using a combination of license plate geometry and color information to locate the license plate area,then from the positional relationship between the vehicle logo and the license plate,according to statistical information to determine the area that the vehicle logo may arise,take this area as the output of coarse positioning.After binarization processing, image were done horizontally, vertically projected, and analysis the projection, according to the statistical properties of the completion,precise positioning the logo. (3)Based on the study of a variety of vehicle logo identify methods, we proposed a vehicle logo recognition algorithm based on Discriminative D-KSVD. This method learning over-complete dictionary from training samples, the test image can be linear combined of multiple atoms of training samples, by solving the coefficient, while increasing the classification items to the K-SVD algorithm, finally finished the target recognition process.Experiments show that, compared to the method based on principal component analysis, and the sparse representation-based representation methods, vehicle logo recognition algorithm based on discriminant K-SVD proposed in this paper is not only reasonable, but also for the presence of noise or vehicle tilt under the circumstances a higher recognition rate.