Allow different smoothing methods

[jni]

Currently, gaussian smoothing is mandated. Two other suggestions by Adam:

• Total variation denoising (already in scikit-image)
• WLHS (paper 1, paper 2)

[jni]

Notes from paper 1:

• cursor width for hysteresis should be 2 times the standard deviation of the noise, and smaller than the difference between background and signal.
• NLM performs almost as well — and is already in scikit-image. ;)
• Having said that the noise images in fig 10 are very impressive.