SAYISAL İMGELER İÇİN PRNU VE CNN TABANLI BÖLGESEL MÜDAHALE TESPİTİ

Abstract

Detecting various forgeries on digital images is becoming more difficult due to the complexity of developing software. As a solution to this complexity, in addition to conventional detection methods, convolutional neural network (CNN) based methods have been developed in recent years. Thus, networks capable of detecting even very complex interventions could be trained. In this paper, a new approach to the convolutional neural network (CNN) based camera model classifier method is compared with the sensor-based PRNU (Photo Response Non Uniformity) method, which is one of the classical methods that can detect local detection using small-scale windows. Thus, which method is more successful is revealed in detail. A total of 26 camera models and the CNN model, which was trained with 96 x 96 pixel blocks selected from these camera models, was compared with the PRNU method using both the 96 and 128 window size. As a result of this comparison, CNN based camera model classifier has been shown to be more successful than PRNU method in the local tamper detection problem.