The tasseled cap result is not appropriate for a confusion matrix, because it contains one band for each category, instead of assigning each pixel to one category. The tasseled cap results are in this form because the different categories are not exclusive. For example, a pixel could have high wetness AND high greenness, etc. A confusion matrix assumes that there is one correct category for each pixel. Regards, Peg

What if i go with 3-layer stack for T-cap (Brightness, greenness and wetness)... then the confusion matrix will take it as 3 different categories for pixels?

No, it would still need to be a classification image, which is a one-band image with each pixel value specifying a different class. So, each pixel has just one class assigned to it. What you could do is run ENVI's Rule Classifier on your Tassled Cap results, and pick whichever of the Tassled Cap categories has the largest value as the "class" that each pixel is assigned to. I don't know if this really makes much sense, though. As I said before, the Tassled Cap categories are not exclusive. So you can have high values in multiple categories. A confusion matrix does not make sense in this context, because it is designed to provide information about how good a classification is. I don't think that running a confusion matrix on a Tassled Cap "classification" generated as described above will tell you much about how valid the Tassled Cap calculation is. - Peg

Confusion matrix do well even in this case, if the components of T-cap are used separately as in comparison with other (i.e. NDVI)