A deep learning approach for facial detection in targeted billboard advertising / Lau Sian En

Lau , Sian En (2025) A deep learning approach for facial detection in targeted billboard advertising / Lau Sian En. Masters thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (22Kb)
	PDF (Thesis M.A) Download (1194Kb)

Abstract

Deep learning has significantly changed industries by facilitating the development of more intelligent and adaptive systems, with applications especially in advertising. Facial detection using deep learning in advertising offers the potential for highly personalised and effective marketing by leveraging real-time consumer demographics. This research explores a deep learning-based facial detection system for targeted advertising, aiming to enhance consumer engagement by delivering personalized advertisements. The research focuses on addressing key difficulties including lack of audience-targeted delivery, real-time implementation challenges and model accuracy difficulties. This system utilises sophisticated deep learning algorithm using Convolutional Neural Network (CNN) to identify and examine human faces, enabling advertisers to customise their content according to demographic variables including age and gender. The system has two modules which are the Realtime Module and the Dataset Evaluation Module. The system employs Multi-task Cascaded Convolutional Networks (MTCNN) for face detection in the DeepFace model, processes webcam photos, predicts age and gender, and maps relevant advertisements accordingly. The evaluation process encompasses real-time performance analysis and testing using the Wikipedia dataset, evaluating the accuracy, precision, recall, F1-score, and confusion matrices. The system’s capacity to provide targeted advertising not only enhances user experience but also greatly enhances consumer engagement. Results indicate that the Realtime Module attains an accuracy of 70% in age prediction and 90% in gender prediction, whereas the Dataset Evaluation Module achieves an accuracy of 74% for age prediction and 90% for gender prediction, hence enhancing advertisement relevance. The study indicates that using facial recognition technologies in advertising tactics can transform conventional advertising methods, providing real-time, adaptive solutions customised for diverse audiences.

Actions (For repository staff only : Login required)