Improved median-filtered data embedding method for image enhancement

This paper proposes an improvement for a data embedding method to embed the data while performing image enhancements via the concept of Median Filter. The pixels within a predetermined window size are gathered and sorted to obtain the partitions of interest for data representation. The centre pixels will be replaced with the to-be-embedded data in a sliding window manner until pixels are replaced. However, the modification of pixel values after the embedding process makes the data extraction process challenging. In this work, various data extraction methods are tested, including repairing using median filter during data extraction in a sliding window manner, and the reverse manner to understand their effects on the image quality and data extraction accuracy. In addition, by implementing majority vote, the data extraction accuracy is significantly improved. The experiment is conducted using the BSDS300 dataset, and it is observed that the image quality can be improved up to 9.5% while the data extraction accuracy is improved up to 7.7%, averagely.