Multi-source motion images fusion based on 3D sparse representation

In order to effectively fusing multi-source images of the same scene, this paper proposes a novel multi-source motion images´ fusion framework based on 3d sparse representation. Compared to the single-frame image fusion technology, the temporal dimension of the motion image is needed to be considered by using sparse coefficient across the adjacent front and rear frames. This helps to obtain a more efficient algorithm and better fusion quality. In addition, the proposed algorithm uses 3D atomic block, make full use of the space-time motion image sequence information, removes redundant dictionary atoms and improves dictionary generation rules to reduce the number of iterations. The proposed fusion framework consists of four steps, i.e. training and updating the dictionary, finding the sparse coefficient, coefficient fusion, image reconstruction. The experimental results demonstrate that, the proposed based on 3D sparse representation fusion method has superior performance to the traditional methods (Dual-Tree Complex Wavelet transform, discrete wavelet transform and Ordinary sparse representation) on objective and subjective metrics.