# How does filter bank work in image processing?

I read a paper about directional filter bank in image processing and I don't understand clearly about the process. It says they use '2-band structure' like this image:

could someone here please explain it to me how this process works? I would like to implement it on python or MATLAB.

• Could you add a reference to the paper please? Just the figure is a bit short. – sansuiso Mar 23 '13 at 22:21
• @sansuiso This is the link to the paper. I hope it can help much. – nsrjws Mar 23 '13 at 22:50
• intuitively speaking; this is about separating approximation and details from the signal and the down sampling it to match the actual number of input samples; here approximation means low frequency and detail means high frequency while Q shows the process of down-sampling. The Hw0 and Hw1 are high and low pass filters – Sufiyan Ghori Mar 24 '13 at 21:31
• @Effected how does the down-sampling process work? Is it a convolution process? and is the modulation by e is also convolution process too? – nsrjws Mar 26 '13 at 0:18
• have a look at what we do when we talk in wavelet domain oi48.tinypic.com/4hruwn.jpg – Sufiyan Ghori Mar 26 '13 at 12:40

1. $H_0$ is a low-pass filter, while $H_1$ is a high-pass filter. You can note that the union of both domains correctly yield the full bandwidth of the input signal.