Logo Recognition Based On The Improved SIFT Algorithm

Vehicle identification need to use the vehicle information as much as possible to identify vehicles on the road, so we require a variety of vehicle identification technology.As a symbolic image of the car,the logo contains the vendor information which is very difficult to be changed, so we must not neglect the role of vehicle-logo recognition in improving the accuracy of the vehicle, and it's a new research direction in vehicle identification technology.Because the logo itself has some special properties such as small targets, high similarity, complicated background and easily affected by the shooting angle and light, which makes the accurate positioning and recognition become a challenging research topic and a big difficulty for vehicle-logo recognition.This dissertation determines the approximate location of a logo in the vehicle image by using the prior knowledge about the relative position between logo and license plate first,and extracts the vehicle-logo rectangular area. And then based on the relative position of logo and the radiator, the logo rectangular area is divided into two categories. For the logo in the centre of the radiator, we use the characteristics that the symmetrical position on the logo left and right sides have the same background texture the same, proposing a method based on template matching and edge detection precision positioning for logo; For the logo above the radiator, we use the characteristics that the texture In the background of logo is simple, and on the edge of the logo is complicated, proposing a method based on edge detection and morphological filtering precision positioning on logo. For the precision positioning logo image, on the basis of traditional SIFT algorithm is presented, this dissertation changes the feature points square neighborhood into circular neighborhood to enhance the global descriptor describes features, so it can make the rotation of the SIFT operator have complete scale invariance. First respectively extracts the feature vectors from the logo template image and the inspected vehicle logo image. Then calculate the Euclidean distance for feature vectors. Using the BBF search algorithm to judge the similarity of two images, it can finally achieved the identification of the logo image effectively.Practical tests show that logo detection and recognition based on the improved SIFT algorithm isn't limited by the logo position and logo background texture. For the change of the image scale and perspective, target obscured, it has good robustness and a strong resistance to rotation scale invariance. So the algorithm can improve the vehicle-logo recognition rate effectively.