AN QUASI INTERPOLATIVE APPORACH OF MLP CLASSIFIER FOR IMAGE RECONSTRUCTION WITH CRYPTANALYSIS

Authors

  • Badde.HariBabu Author
  • Dr. Vikas Kumar Author
  • Badde. Srinivasa Rao Author

Keywords:

Cryptanalysis, Deciphering, Multilayer Perceptron (MLP), Encryption

Abstract

In the realm of image reconstruction and
cryptanalysis, the development of accurate and
efficient classifiers plays a pivotal role in deciphering
encrypted images and restoring them to their original
form. In this study, we propose an innovative
interpolative approach of a multilayer perceptron
(MLP) classifier tailored specifically for image
reconstruction tasks involving cryptanalysis. Our
method aims to achieve superior performance in both
decrypting encrypted images and accurately
reconstructing them, leveraging the flexibility and
adaptability of MLPs in learning complex patterns and
relationships within the data.
The proposed interpolative MLP classifier is designed
to operate in two distinct modes: one for processing
real (unencrypted) images and another for decrypting
encrypted images. In the real image mode, the
classifier learns to recognize and classify various
features and patterns present in unencrypted images,
achieving a training accuracy of 99% through rigorous
training on a diverse dataset. This high level of
accuracy ensures the reliable reconstruction of real
images, enabling the classifier to faithfully restore
images even in the presence of noise or distortion.
In the cryptanalysis mode, the interpolative MLP
classifier is tasked with decrypting encrypted images
by discerning hidden patterns and structures obscured
by encryption algorithms. Through an innovative
interpolation technique, the classifier adapts its
learned representations from real images to decrypt
encrypted counterparts, effectively bridging the gap
between the two domains. By exploiting the inherent
similarities between real and encrypted images, the
classifier achieves remarkable performance in image
reconstruction, with decryption accuracy matching or
exceeding that of conventional cryptanalysis methods.
Key to the success of our approach is the custom
design of the MLP architecture, tailored to
accommodate the unique characteristics of image data
and encryption schemes. The network architecture is
optimized to handle high-dimensional image data
while incorporating advanced activation functions and
regularization techniques to prevent overfitting and
enhance generalization. Additionally, the training
process is augmented with specialized loss functions
and optimization strategies tailored specifically for
image reconstruction and cryptanalysis tasks, ensuring
robust convergence and optimal performance.
Through extensive experimentation and evaluation on
benchmark datasets, our interpolative MLP classifier
demonstrates superior performance compared to
existing approaches in both image reconstruction and
cryptanalysis. With a training accuracy of 99% in both
real and decrypted cases, our method establishes a new
benchmark for accuracy and efficiency in image
reconstruction tasks involving cryptanalysis, offering
promising implications for various applications in
security, forensics, and image processing.

Downloads

Published

01-04-2024

How to Cite

AN QUASI INTERPOLATIVE APPORACH OF MLP CLASSIFIER FOR IMAGE RECONSTRUCTION WITH CRYPTANALYSIS. (2024). International Journal of Engineering Research and Science & Technology, 20(2), 810-821. https://ijerst.org/index.php/ijerst/article/view/344