Tried to implement a paper based on its abstract, because I'm not willing to pay 40 USD. The result is a mix of success and failure.
The paper introduced a steganography called DWT-DCT-SVD, which is applying DWT to a picture and then applying DCT and finally encoding bits into the S matrix of the SVD operation.
The first picture is encoded picture after jpeg compression (photoshop, quality 8, raw picture is 90MB in png), picture 2 is the intended decoded result, and picture 3 is the jpeg decoded result. I would say not bad...
... unless you zoom in on the picture and notice there are a lot of stripes giving off the fact that this picture is suspicious (picture 4).
To resist jpeg compression, I use the LL matrix of the Haar DWT result (which should contain low freq data that can be preserved by JPEG), then chunk it into 4x4 chunks, then apply DCT and SVD to each chunk, then encode the bits into the first and second element of the S matrix.
It certainly works, but any other steganography could work too if it leaves a strong and visible mark on the image, which surely can be preserved by JPEG compression (Modify low freq in FFT space, or just use the MSB).
I swear there are visible stripes in the dark area, like the screenshot I take in picture 4. But I don't know why it's not visible anymore, probably because of the compression. LOL