Deep learning-based biometric image feature extraction for securing medical images through data hiding and joint encryption-compression

<p>Images are promising information carriers when compared to other media documents in the healthcare domain. However, digital data transmission over unprotected wired or wireless networks poses a threat to the security of healthcare systems. As a result, the issue of copyright violation and identity theft can occur due to the unauthorised use of these data. This paper proposes a new secure method under a framework that embeds <a href="https://www.sciencedirect.com/topics/computer-science/biometrics" target="_blank">biometric</a> fingerprint <a href="https://www.sciencedirect.com/topics/computer-science/image-feature" target="_blank">image features</a> in a medical image without any <a href="https://www.sciencedirect.com/topics/computer-science/perceptual-distortion" target="_blank">perceptual distortion</a>. This paper uses ResNet152 for biometric image feature extraction in the first stage and features to generate a secret key for embedding in the second stage. The method combines encryption and compression scheme based on a generated key, novel <a href="https://www.sciencedirect.com/topics/computer-science/chaotic-map" target="_blank">chaotic map</a> and <a href="https://www.sciencedirect.com/topics/computer-science/huffman-coding" target="_blank">Huffman coding</a> to enhance the security of medical images while reducing the storage consumption or <a href="https://www.sciencedirect.com/topics/computer-science/bandwidth-requirement" target="_blank">bandwidth requirements</a> if images are transmitted to remote servers. Experimental results show that the proposed method presents superior security with high <a href="https://www.sciencedirect.com/topics/computer-science/imperceptibility" target="_blank">imperceptibility</a> and <a href="https://www.sciencedirect.com/topics/computer-science/compression-performance" target="_blank">compression performance</a>, ensuring its effectiveness as an image protection mechanism for medical applications. Extensive experimental results show that the proposed method achieves an average peak signal-to-noise ratio (PSNR) that is above 54 dB, a structural similarity index measure (SSIM) close to 1, a bit error rate (BER) of 0 and a normalised correlation (NC) of 1. Moreover, this method compresses the images up to 70% when tested on three standard datasets. </p>