Hi Qing Li,
I understand that extracted watermark in Experiment 2 is more robust to JPEG compression compared to Experiment 1 can be attributed to several factors related to the properties of the LL2 subband in a 2-level Discrete Wavelet Transform (DWT) decomposition:
In Terms of Frequency Content:
Low-frequency components typically contain the most significant parts of the image, such as background and major structures, which are less likely to be altered by compression and other common image processing operations.
- LL2 subband: This subband contains the lowest frequency components of the original image. JPEG compression primarily affects high-frequency details, leaving low-frequency information largely intact. By embedding the watermark in the LL2 subband, you are placing it in a region less susceptible to modification by compression artifacts.
- LL subband: This subband still contains some higher-frequency details compared to LL2. While it remains robust to some degree, these higher frequencies are more vulnerable to compression artifacts, potentially affecting the embedded watermark and leading to higher Hamming distances.
In Terms of JPEG Compression:
- JPEG compression involves quantization, which discards some information to reduce file size. The quantization step typically affects high-frequency components more severely than low-frequency ones. Since the LL2 subband primarily contains low-frequency information, the embedded watermark undergoes less severe quantization, mitigating its distortion and improving robustness.
Overall, by embedding the watermark in the LL2 subband, you leverage its properties of low-frequency content, reduced quantization impact, and concentrated energy to achieve greater robustness against JPEG compression compared to the LL subband in Experiment 1.
I hope it will be useful!
