Enhanced Mayfly Optimization Algorithms with Pixel-Based Learning Color Space Data Hiding Using a CNN-LSTM Network
J. Jude Moses Anto Devakanth, R. Roselinkiruba, D. Arul Suresh, N. Naveen Kumar, C. Saranya Jothi, S. Balamurugan · IETE Journal of Research · 2025
In this research, a novel Mayfly Optimization with Pixel-Based Learning (MOPBL) framework is proposed to enhance the performance of Convolutional Neural Network – Long Short-Term Memory (CNN-LSTM) networks for optimal pixel selection in data-hiding applications. Traditional Mayfly Optimization Algorithms (MOA) suffer from premature convergence and susceptibility to local optima. To overcome these limitations, the proposed MOPBL method introduces a rank-based system that searches for pixels similar to the secret data within the RGB color space using histogram analysis. The method employs four strategic embedding schemes across RGB channels to identify optimal pixel candidates. Once identified, each color channel is partitioned into 3 × 3 blocks, and the Complexity of the Block (CB) is assessed. Blocks with low complexity are further refined using a Clustering Algorithm (CA) to isolate the Region of Interest (RoI). Additionally, Prediction Error (PE) is evaluated to minimize embedding distortion. The final data embedding is performed within the selected, high-quality pixel regions. Experimental results confirm that the proposed MOPBL-CNN-LSTM integration significantly improves embedding performance in terms of PSNR, embedding capacity, and robustness. Moreover, the method demonstrates strong resilience against steganalysis attacks such as RS analysis and chi-square tests.