Abstract:
Nowadays, there are many types of image media on the Internet which anyone can easily
access to inappropriate media. The problem is that it is difficult to protect young children from
viewing pornographic images. For this reason, this master project proposes to use an artificial
neural network (ANN) model to classify the pornographic images. The research procedures are as
follows: 1) data preparation, 2) feature extraction, 3) artificial neural network construction, and 4)
model evaluation. In the construction process, there were 10 ANN models; each model was
trained 10 times to select the best design. In verifying process, the efficiency of ANN models
were considered by using mean squared error (MSE) based on both training data and test data.
The best model was selected from the lowest MSE. For the results, ANN with the design 6 inputs,
4 hidden nodes, and 1 output (6-4-1 model) is the best model with the value of MSE of
3.718× 10- 5 on training data and 3.2×10- 3 on the test data. The number of misclassified
images was 12 pictures from 400 images (error 3.00%). That means the ANN 6-4-1 model is a
suitable design to be further developed effective classification software for pornographic image
detection.