Abstract:

A novel image fusion scheme based on quaternion wavelet transform and sparse representation (QWTS) is proposed. This method complements the defects existing in traditional multi-scale analysis based image fusion and sparse representation based image fusion. The proposed QWTS based image fusion scheme performs the three stages. Firstly, the quaternion wavelet multiscale decomposition of each source image is performed to obtain their low-pass and high-pass coefficients.Secondly, the fused quaternion wavelet coefficients are obtained in the low-pass bands with a sparse representation based fusion approach while in the high-pass bands by fusing the largest absolution values. Thirdly, the fused image is reconstructed by taking inverse quaternion wavelet transform. Furthermore, the proposed image fusion scheme is analyzed theoretically. The performance of the proposed image fusion scheme is evaluated on six groups testing images, and the fused results is compared with those results obtained by the sparse representation, the discrete wavelet transform, the dualtree complex wavelet transform and the quaternion wavelet transform image fusion schemes subjectively and objectively. Experimental results show that the proposed fusion scheme is very effective.