Integrating Image Steganography Techniques Into an Existing Efficient Image Recognition Artificial Neural Network for Optimal Performance

Abstract

Image steganography is a data-hiding process whereby a digital image media is used as a cover to encrypt and embed data within the pixels of the image. This method can be applied in various research fields, including image processing, power line systems, mobile applications, etc. This paper investigates the application of an image steganography technique that differs from traditional least significant bit substitution methods. The discrete cosine transform least significant bit-2 method is proposed in this paper. Rather than placing the hidden data bits in a deterministic manner, the proposed method inserts the data bits in random pixel locations within the cover image to alleviate steganalysis and unauthorized access to the hidden data. An image recognition artificial neural network developed using open source TensorFlow is applied and used to test the effectiveness of the prosed encryption method. The proposed encoding scheme shows more robustness and allows for higher neural network prediction confidence. This method can be applied in power line systems to determine the physical conditions of power lines, such as when they are damaged/broken. The image recognition artificial neural network can also forecast power line maintenance cycles for proactive monitoring of power systems.