Image Fusion

Hi

I am working on empirical mode decomposition and my objective is image fusion. I have intrinsic mode functions for two images and i want to fuse it. i have applied Principle component analysis and got better results when compared image fusion without applying EMD. But when i apply discrete wavelet transform it is giving good results when number of wavelets is less than three. but when i increase the number of wavelets more than that it is not giving good results when compared with the former. Is it a concept that if i decompose the already decomposed images i will be getting bad results eh??Experts please help..