Research On Image Fusion Method Based On Regional Characteristics In Multi-scale Domain

As one of the indispensable technologies in the field of image processing and computer vision,image fusion can obtain judgment results more accurately than a single information source.In order to obtain a more sophisticated and efficient fusion method,this paper has studied a large number of existing theories and achievements in the fields of fusion.Using multi-focus images and multi-modal medical images as the research object,the analysis and research are carried out in the multi-scale transform domain.The main work and innovation points of this paper are as follows:This paper proposes a new fusion method based on compressed sensing in multi-scale domain—Image fusion method based on NSST and CS combined with regional characteristics.After the decomposition of NSST,the high frequency component coefficients with high sparsity are compressed by gaussian measurement matrix.Before using the orthogonal matching tracking method,the measurement values are fused in the largest absolute.Considering that there is no high sparse property in low frequency component coefficients,the fusion rules of region energy and regional variance-weighted adaptive is used directly to complete the fusion,and then fusion results obtained by NSST inverse transform are used to perform the comparison and evaluation in the hope of effectively improve the image fusion.Besides,with the help of CS,the fusion speed has significantly improved.An image fusion method based on regional characteristics in multi-scale domain is proposed.In order to solve the problem that there is no translation invariance in general multi-scale transformation,and the multi-scale decomposition of NSCT often leads to the problem of image frequency band aliasing,a new transform(àtrous – NSCT)is introduced.Through experimental data analysis,the optimal multiscale decomposition layer of this transformation is 4.Considering the influence of the neighborhood characteristic information for image information,using regional variance as active measurement to analyze the high frequency coefficient,and then regional variance-based weighted adaptive model is adopted to the fusion.As for the approximate information contained in the low frequency coefficient,the regional average gradient is used as the measurement factor,and the fusion rule based on the regional average gradient maximum is chosen to start the fusion operation.Finally,the high frequency fusion operator and low frequency fusion operator are reconstructed by àtrous wavelet inverse transform.Comparing with the other five methods from the aspects of the subjective visual and objective data,the fusion effect of this method is better than the mentioned algorithms.