A classifier-free codebook-based image classification of vehicle logos

摘要: Logo identification and classification have received considerable attention from both the machine learning computer vision communities. Vehicle logo recognition (VLR) is used to recognise accurately manufacturer of a vehicle by using its iconic logo. A VLR system in addition license plate aims increase confidence monitoring systems private environments such as car parks companies, shopping malls, institutions. challenging process due presence extensive background, clutter, different degree illumination, varying sizes vehicles motion, change weather conditions fog, sunny rainy which are present two-dimensional images. On other hand, bag-of-features (BoF) approach initiated be black box providing reliable repeatable measurements images for wide range applications visual object recognition. The advantages BoF simplicity state-of-the-art performance recent tasks. Most literature has focused on less than twenty distinctive logos vehicles. In this paper, we propose novel an alternative method extracted patch-based descriptors test image voted against locally merged codebooks predict class label without need mapping into fixed-length feature vector then feeding it standard classifier. proposed SIFT compared with traditional that employs classifier: nearest neighbour support machines. evaluated 25 classes 20 per class. Testing results show our promotes deliver robust 98.6% drastically reduces time needed approach.