After testing our algorithm on three different images we noticed that the accuracy of extraction corresponded to the characteristics of the particular images.
We hid a message in these three different pictures:
Picture #1: Phil "A" |
Picture #2: Gas Leak |
Picture #3: Phil "B" |
The following table displays our results:
Picture # | Characteristic of Image | Output | Accuracy |
1 | High Frequency | rich b.3in 2004! | 99% |
2 | Middle Frequency | rich c. ?n 20044 | 97% |
3 | Low Frequency | /9c 39 im 0047 | 84% |
These results showed us that our algorithm was substantial and worked effectively for some pictures. As we looked more closely at the images that hid data more effectively than others we realized that there was a direct proportionality between bit accuracy and frequency content of an image. Our first image, which worked very well with our algorithm, has significant high frequency components in the water on the bottom left and the tree. However, our last image had very few high frequency components due to the plain background and thus did not work as well.