Improving data extraction accuracy for median filter-based data hiding method

Abstract

In this paper, an improved method of the median filter-based data hiding method is proposed. The median filter can be performed while performing embedding on images, to achieve both image enhancement and data embedding in one step. However, the low data extraction accuracy in the existing median filter-based data hiding method is a concern. Along with that, the triple embedding, repairing, reverse scan order, and majority voting approaches are incorporated into the embedding and extraction process. This helps improve the accuracy of the median filter-based data hiding method while ensuring image enhancement on both noisy and non-noisy images. In this work, different experiments are conducted using various settings of noise types, noise levels, image window sizes, subsets, and pixel-pair to evaluate the performance of the approaches. The result shows an overall improvement in accuracy when triple embedding for data embedding, reverse and repairing with majority voting for extraction is performed. In terms of image quality, both the reverse-repair and majority reverse-repair methods exhibit significant improvements during data extraction, especially when removing the Salt&Pepper noise and Speckle Noise. In the best case, 100% extraction accuracy can be achieved when the window size is 5 × 5 and 7 × 7.