Hybrid plugin for detecting illicit images on the internet using EfficientNet convolutional neural networks

Christine Laure Mananga (University of Douala)
Felix Paune (University of Douala)
Léandre Nneme Nneme (University of Douala)

Indonesian Journal of Electrical Engineering and Computer Science
Vol 42, No 3: June 2026

The proliferation of illicit visual content on the internet, such as pornography and violent imagery, presents a growing societal concern. This paper proposes the design and implementation of a lightweight browser-integrated plugin that utilises a hybrid approach combining content based filtering with convolutional neural networks (CNNs), specifically the EfficientNetB7 architecture, to detect and block illicit images in real time. Developed using Python and TensorFlow, the plugin was trained on a curated dataset comprising NSFW, DeepNude, and safe-for-work (SFW) images. Experimental results on a dataset of 1,064 randomly selected images demonstrated a detection accuracy of 99%, with a processing time of 92 seconds and a 7% combined false positive and false negative rate. The plugin is compatible with Chrome browsers and contributes to safer online experiences, particularly for children, educators, and users in sensitive environments.