Detection and Classification of Logos and Trademarks from Images

In the field of image analysis, various techniques are employed to extract and interpret relevant information from images, such as logos or recurring patterns. Object detection and classification play crucial roles in this process. While the human brain can locate, detect, and recognize an object in images within minutes, machines require substantial time and extensive data to perform the same task. This paper proposes the use of transfer learning as a technique to reduce the time and resources needed. A comparative study of different deep learning-based logo detection algorithms, including Faster RCNN, Mask RCNN, RetinaNet, and Cascade Mask RCNN is presented. The focus is on their application in detecting and localizing brand logos in images from a TV show. The simulations are conducted using the Open Logo challenge dataset, which is widely recognized in this area and comprises 27,000 images from 352 distinct logo classes. The proposed algorithms’ performance in detecting and localizing brand logos is evaluated, and the results are reported in terms of precision, recall, and execution time.