CLASSIFICATION AND DETECTION OF DISTORTED IMAGES WITH DEEP CONVOLUTIONAL NEURAL NETWORK

Abstract

The central theme of this thesis is Machine Learning and its possible application in the field of images and videos. The final aim will be to evaluate the behavior of neural networks as function of the intrinsic quality of the images, starting from the original ones to finish with images that suffer distortions of various kinds. Everything will be carried out by testing and developing different possible applications of artificial intelligence on images and videos having in common denominator the prediction and recognition of the contents. To get a full picture of the state of the art in this field a priori study has been carried out on what is this field, and about what it may be in the future, in terms of technologies and mathematics approache. Specifically, the work carried out is divided into 3 macro sections: a first part in which First a generic in-depth study on Machine Learning was done, then a little more in-depth study on Convolutional Neural Networks (CNN), also providing some mathematical hints behind; in this preliminary phase, the state of the art was also analyzed of neural intelligence in the field of images and videos. The second section of the research was that concerning the development phase of study in the which we tried to build scripts and programs that would allow to exploit what was analyzed in the previous phase. The final phase, that of testing, the central core of this research is the one in which we have been studied and evaluated the behavior of neural networks as function of images given in varies form: distorted images in different ways and with different levels of application with same distortion.