Variational Fusion Strategy for Multiscale Analysis Domain

Ning MA,Ze-ming ZHOU,Li-min LUO
DOI: https://doi.org/10.3969/j.issn.1002-137X.2013.09.060
2013-01-01
Computer Science
Abstract:Conventional fusion strategies typically manipulate the fused coefficiented too directly,which may introduce artifacts and decrease the fused image quality.This paper presented a variational fusion strategy for multiscale analysis domain.An energy minimization problem was proposed to find the solution whose gradient is closest to that of the target gradient field,which is built from the gradient information of the source images' decomposition coefficients.To keep the similarity with source images,the constraint condition was imposed into the highpass and lowpass coefficients separately.Since the highpass coefficients are sensitive to the noise,the total variation term is also introduced to assure the smoothness of the solution.The performance of the proposed fusion strategy was evaluated on multifocus images,infrared and visible images.Experimental results show that the strategy can inject more spatial detail information and meanwhile preserve more structural information of the original image than conventional fusion strategies.
What problem does this paper attempt to address?