Towards a Novel Cryptanalysis Platform based Regions Of Interest Detection via Deep Learning models

Cryptanalysis is an audit step that leads designers to develop more robust cryptographic algorithms and assess algorithms’ overall performance. The fundamental problem is that the human evaluation of the cryptanalysis results is essential in this process. It optionally allows the remarkable convergence towards a promising result if it is based on better criteria, as it does not allow to find any solutions.To overcome the human intervention in this process we propose, in this work, a new cryptanalysis platform of image permutation-only cipher based on the detection of significant parts (ROIs) implementing the genetic algorithm and two models based on deep learning namely: Faster R-cnn for object detection and Mask R-cnn for segmentation.This is to automate the process of decryption keys evaluation and minimize the search space, which makes it possible to directly determine the permutation key or the most part of it. This work is applicable to color (RGB) images encrypted by pixel permutation techniques. It is independent of the permutation algorithm and it based on cipher text only attack by the advantages of those models exploitation to discover the correlation between adjacent pixels and to ameliorate this significant correlation by the genetic algorithm.