SELECTED NIGERIA FEDERAL UNIVERSITY CLASSIFICATION BY LOGO WITH MAXIMALLY STABLE EXTREMAL REGION (MSER) AND CONVOLUTIONAL NEURAL NETWORK (CNN)

Abstract

Logo recognition is important in various applications, including branding analysis, advertisement monitoring, and image retrieval. This paper presents an approach for the classification of selected logos of Federal Universities in Nigeria, leveraging the combination of Maximally Stable Extremal Regions (MSER) and Convolutional Neural Networks (CNN). The proposed methodology consists of two main stages: image region extraction using MSER and logo classification using CNN. In the first stage, the MSER algorithm was employed to detect stable and distinctive regions in the input images. This approach effectively captured the salient features of the logos. These MSER regions were bounded by polygon and preprocessed to generate a dataset for training a CNN model. In the second stage, a deep learning architecture based on CNN was designed and trained to classify the logos into the specified university classes. The model's performance was evaluated using accuracy, precision, recall, and F1-score. Experimental results demonstrated that the combination of MSER and CNN recorded 0.9626, 0.9561, 0.9622, and 0.9620 for accuracy, precision, recall, and F1 score respectively. The proposed logo recognition system was compared with two state-of-the-art models and recorded excellent performances. The model provides an efficient and effective solution for identifying and classifying logos in real-world scenarios. It offers potential applications in logo detection, logo-based image retrieval, and brand monitoring, facilitating various domains such as marketing analysis, copyright enforcement, and educational research. This approach's performance indicates its potential for further development and integration into practical systems.