Abstract
The construction and development of new techniques for a nonlinear multivalued Boolean function is one of the important aspects of modern ciphers. These multivalued Boolean functions need to be defined over various algebraic structures which map multiple inputs on multiple outputs. Modern block ciphers are a combination of linear and nonlinear functions which adds diffusion and confusion capabilities. We have offered an innovative system for the construction of the confusion component of block cipher by using recurrent neural networks. Since the confusion component is a multivalued Boolean function, therefore, we need many to many types of recurrent networks with an equal number of inputs and outputs. With this scheme, we have achieved a standard benchmark nonlinear of 112 with balancednesss having low linear and differential probabilities. We evaluated some common and advanced measures for the eminence of randomness and cryptanalytics to observe the efficiency of the proposed methodology. These outcomes validated the generated nonlinear confusion components are effective for block ciphers and have better cryptographic strength in image encryption with a high signal-to-noise ratio in comparison to state-of-the-art techniques.