Recognizing cars is a highly difficult task due to the wide variety in the appearance of cars from the same car manufacturer. Therefore, the car logo is the most prominent indicator of the car manufacturer. The captured logo image suffers from several problems, such as a complex background, differences in size and shape, the appearance of noise, and lighting circumstances. To solve these problems, this paper presents an effective technique for extracting and recognizing a logo that identifies a car. Our proposed method includes four stages: First, we apply the k-medoids clustering method to extract the logo and remove the background and noise. Secondly, the logo image is converted to grayscale and also converted to a binary image using Otsu's method. Thirdly, the Daubechies wavelet with DCT transforms is applied to extract a feature vector for each image. Finally, the Canberra distance is used to match the tested image's feature vector to all feature vectors in the database. The test results indicate the highest CRR, accuracy, and precision at 99.37%, 99.39%, and 99.80%, respectively. This system is applicable to intelligent surveillance systems.
