Abstract

The use of digital media for advertisement is on the increase due to advancements in technology and mobile devices and in their applications. Prominent of the digital media is images. Digital Advertisement images are images of products sold and/or services rendered by the advertiser. In addition, advertisement images also contain textual information that helps connect the potential customer to the advertiser, such as telephone numbers. Potential customers tend to copy or cram the contact information which could lead to the loss of this information. The incorrect information. will hinder the aim of the advertisement from being fulfilled. The Automatic detection of these texts using computer vision algorithms can foster the establishment of communication between the advertiser and the potential customer. In this study, the You Only Look Once (YOLO) model is utilized for text detection from advertisement images. YOLO is a widely used object detection algorithm in particular YOLOv5. It is known for its balance between speed and accuracy of detections and has been used in various sectors. However, its performance for text detection from advertisement images is yet to be explored. Therefore, this paper investigates and compares the performance of various variants of YOLOv5 for the accurate detection of texts from digital advertisement images. Results of experiments showed that Yolov5x achieved a precision, recall, and mAP50 of 98.9, 97.7, and 99.4.