Comparison of 3-D and 2-D DCT Based Filtering of Multichannel Images

A common feature of multichannel images of different origin and spectral bands is that they exhibit a high degree of mutual correlation of their components. At the same time, information content and noise properties in component images are more or less different. For the noise removal from such images, two general approaches are possible. The first one assumes component-wise (2-D) filtering; the second approach presumes vector (3-D) processing. Component wise filtering is, in general, simpler whilst 3-D processing that is able to take into account the inter-channel (inter-band) correlation of data is potentially more efficient.