Common Small Vehicle-logo Recognition

At present, although the key technology of the Intelligent Transportation System (ITS) is License Plate Recognition, with the help of vehicle-logo recognition we could get vehicle information more accurately. These information would provide lots of favorable evidences about vehicle crime. In this paper, the technology of vehicle logo recognition has carried on research and discussion.In the stage of vehicle logo location, we find the logo based on the location of vehicle license. Using the relationship between logo and license we could get a region which contains the logo, and then remove the noise and stretch gray level. With the minimum filter and Gaussian filter, we could estimate the background light conditions and then we remove it, to reduce the affect of the uneven illumination. Then use the horizon projection of the edge image to make sure the up and down border.In the stage of vehicle logo enhancement, as the MultiScale Retinex has shown the best effect, we decide to use this method to enhance the logo image. Use Sobel Operator and Shen Operator getting the logo edge image and use the vertical-projection of the logo edge image for calculating the axis of symmetry.In this paper, we thought the size and the shape as an important feature of vehicle logo, so we build7kinds of size templates such as32*32,48*48,56*32, to create logo library which include44kinds of cars. We use gradient feature to describe the logo and use template matching method and Nearest Neighbor Classifier to recognize the logo. The first five results are considered as the final results. In the high quality images case, the recognition rate could be93.71%. In the low quality images case, the recognition rate could be88.4%.