Video Steganography Algorithm Defining the ROI using Haar-Like Features and Convolutional Neural Network

For centuries, humans have sought methods to conceal information and transport it from one location to another. In modern times, computers are utilized to conceal and transport digital data. The process of hiding a secret message inside another ordinary digital data, such as a digital image, to prevent the detection of the secret message is referred to as steganography [1]. Many algorithms have been developed to embed secret messages, with undetectability being one of the primary goals. Algorithms such as Kanade-Lucas-Tomasi (KLT) and Discrete Wavelet Transform (DWT) are utilized within algorithms to determine Region-of-Interest (ROI). This paper proposes a new approach for determining the ROI by replacing KLT or DWT with a Convolutional Neural Network (CNN). In each frame, windows with face-like impressions will be determined using Haar-like features (HLF). Each window is then passed to a CNN trained on one particular person, referred to as the person-of-interest. The system will recognize if the person of interest is present in the frame and use that window as the ROI.