SCRAMBLED IMAGE ENCRYPTION ALGORITHM BASED ON NEURAL NETWORK

As a traditional communication method, image is widely used because it is easier to process than text. However, the user's communication environment may be very dangerous. In this case, the security of information becomes crucial. Therefore, the encryption method used to convert the image into another image that is not easy to understand is very important to maintain the security of the transmitted information. The primary objective of this paper is to introduce a novel approach for image encryption that leverages machine learning and cryptographic techniques. This method employs pixel substitution and convolutional neural network (CNN) technology to effectively conceal information in a manner that is both non-uniform and efficient, thereby enhancing the security of image data transmission. It has been verified by experiments that this method of image encryption has a good effect throughout the process of secure encoding and subsequent decoding. The NPCR value after decoding is 100%, and the restoration degree of the decrypted image is 92%, which effectively retains the original image information. It is more effective than traditional chaotic system encryption methods.