Change the method of removing spikes on AutoDecoder

[electricalgorithm]

AutoDecoder works on the principle by substracting the average of the signal from itself. Since encoder puts the secret images on top of the frequency magnitude within an offset, this method works as expected. However, it is easily seen that it is not a bulletproof and causes some problems.

Bug Description: After choosing a carrier image at X size in vertical, and configuring the secret images above or equal to (X-300) size, it causes some pixels to be eliminated on the decoding side.

Ideas to Solve: There must be a statistical mathematics where we can eliminate these big spikes on the edges. Maybe after using fftshift, this spikes will be collected on the middle. We can apply a low-pass filter to eliminate them, but need to control the change in secret images' magnitudes.