Multimodal image fusion via sparse representation and clustering-based dictionary learning algorithm in NonSubsampled Contourlet domain

Image fusion is a widely used technique for enhancing the interpretation quality of images in medical application, which use different medical imaging sensors. This paper presents an image fusion framework for images acquired from two distinct medical imaging sensor modalities (i.e. PET and MRI) based on sparse representation in Non Sub-sampled Contourlet transform (NSCT) domain. NSCT firstly performed on pre-registered source images to obtain their low-pass and high-pass sub-bands. Then, low-pass sub-bands are fused by sparse representation based approach, using a clustering-based dictionary learning while high-pass sub-bands are merged using salience match measure rule. Constructing a compact and informative dictionary is an important step toward a successful image fusion technique in sparsity based models. This paper presents efficient dictionary learning method by clustering patches of several directional sub-bands of different source image in NSCT domain. The experimental results demonstrate effectiveness of the proposed dictionary learning algorithm and priority of general framework, in terms of both subjective and objective evaluation for image fusion task.