Deep Active Learning for Pornography Recognition


Hor, Sui Lyn and Abdul Karim, Hezerul and AlDahoul, Nouar and Lye Abdullah, Mohd Haris and Mansor, Sarina and Ahmad Fauzi, Mohammad Faizal (2022) Deep Active Learning for Pornography Recognition. None. (Unpublished)

[img] Other (Presented at SASSP IPIARTI 2022)
MMUPaper2 SASSP 2022 Hor Sui Lyn.pdf - Accepted Version
Restricted to Repository staff only

Download (338kB)


Usages of deep learning networks for pornographic content detection has been explored to an extent that performance achieved is reaching saturation. Bottleneck in labelled data requirement could be overcome through utilisation of active learning framework for deep model training, which the effectiveness has been proven in several other applications. Comparison of classification accuracy and labelled data requirement among deep active learning with sample selection strategies (random sampling, least confidence sampling and entropy sampling) and fully supervised learning method was performed to demonstrate the effects of the proposed training framework. Convolutional neural network named ResNet50V2 and several pornographic image datasets of different image types were utilised in the experiments.

Item Type: Other
Uncontrolled Keywords: Convolutional neural network, deep active learning, pornography recognition.
Subjects: Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions: Faculty of Engineering (FOE)
Depositing User: Dr. Sarina Mansor
Date Deposited: 30 Nov 2022 04:16
Last Modified: 22 Mar 2023 05:01


Downloads per month over past year

View ItemEdit (login required)