Abstract:

Aiming at the problem of multimodality images fusion, we proposed an image fusion method based on boundconstrained optimal projection gradient for nonnegative matrix factorization (NMF) in translation invariance nonseparable shearlet transform (TINST) domain. The problem of low fusion accuracy in some existing typical fusion methods was solved effectively. Images were decomposed to some subbands by translation invariance nonseparable shearlet transform. The lowfrequency subband coefficients were regarded as original observed data, and the lowfrequency subband coefficients were obtained by boundconstrained optimal projection gradient for NMF algorithm. Highfrequency directional suband coefficients were used as external input excitation and edge intensity was served as linking strength of each neuron in pulse coupled neural networks (PCNN) and after the fire processing and compareselection computing, fused highfrequency directional suband coefficients were obtained. Finally, all the fused subbands were reconstructed to an image by translation invariance nonseparable shearlet inverse transform. In order to verify the efficiency of the proposed method, some compared fusion experiments were implemented on several sets of different modality images. Subjective and objective evaluation on fused image indicates that the proposed method is better than a few existing typical fusion techniques based on multiscale decomposition (MSD).

Key words: translation invariance nonseparable shearlet transform, fusion rule, nonnegative matrix factorization, pulse coupled neural network