Classification of images containing sensitive and non-sensitive information using the convolutional neural network Available to Purchase

Currently, social media is used by almost all ages, from pre-teens to the elderly. They use social media to socialize and express their activities by uploading pictures; however, they sometimes do not realize that the images they upload contain sensitive information. They do not carefully analyze the images to be uploaded due to a lack of knowledge, irresponsible acts, or impaired vision. Therefore, we propose a method to classify images with sensitive or non-sensitive content using Convolutional Neural Network (CNN). This research was performed through several steps: image public dataset collection, data pre-processing, model architecture design, model training, and model validation. A randomly selected sample of 2,000 of 5,537 images from the VizWiz-Priv dataset was used to train the classification model. The CNN architecture was compiled using two max-pooling layers and four convolution layers. Finally, the model was trained and validated using images containing sensitive and non-sensitive information. The results revealed that the model accuracy during training and validation achieved 98.75% and 83.30%, respectively.