Research On The Image Fusion Algorithms Based On Multiscale Transform

Image fusion based on multiscale transform has been the researching focus currently. At first, the feasibility and the developmental history about the application of the Multiscale transform in the field of image fusion are generalized in this paper, and the principles of wavelet and Contourlet are introduced in detail as well as the advantages and disadvantages of this two kinds of multiscale transform are anarlzed and compared. Then, according to theoretical analysis and experimental statistics based on the discussion and the research of the distribution of coefficients after the two kinds of multiscale transform, the regional feature and the autocorrelation modal with regard to low-frequency wavelet coefficients are proposed as well as the directional distribution characteristic and quadtree structure with regard to high-frequency wavelet coefficients are proposed in this paper, and for the directional characteristic of the Contourlet coefficients, the directional modal is proposed. At last, on the basis of distribution model of coefficients, three kinds of image fusion methods based on multiscale transform are proposed as below: (1) Image fusion method based on directional characteristic of wavelet coefficients. For every coefficient of the low-frequency subbands, the algorithm of this paper makes out the final coefficient by using the neighboring relation because of the neighboring relational distribution of low-frequency. And for every coefficient of the high-frequency, the algorithm of this paper respectively adopts the rule such as the directional spread of differences , the relativity of directional field and the energy matching of the directional field. Compared with the traditional rules of image fusion, this algorithm can capture the edge information of wavelet coefficients very well , and both the Entropy and Average grads used to judge the results of the proposed algorithms improve as a certain extent. (2) Image fusion method based on the autocorrelation characteristic of wavelet coefficients. According to autocorrelation characteristics of low-frequency wavelet coefficients, for every coefficient of the low-frequency, the algorithm of this paper makes out the final coefficient by using the covariance relative, and for every coefficient of the high-frequency, the algorithm of this paper adopts fusion rule of the fusing coefficient by using the directional variance coefficients, neighborhood matching coefficients and quadtree matching coefficients. Compared with the traditional image fusion,this algorithm can be sure of fusion coefficients more accurately,and Entropy , Average grads, PSNR and RMSE used to judge the results of the proposed algorithms will improve to a certain extent. (3) Image fusion method based on the directional characteristic of the Contourlet coefficients. For every low-frequency coefficient of Contourlet transform, the algorithm of this paper makes out the final coefficients according to their covariance relative. For every high-frequency coefficient of Contourlet transform, the algorithm of this paper makes out the final coefficients by using the energy maching measure of the directional domain. Compared with the wavelet transform, Contourlet can not only capture edge points and the smoothness along the contour, but also adopts the directional characteristic well. The fusion algorithms mentioned above are used to fuse multi-focus images and medical images and the experimental results show that the proposed methods are more outstanding than the conventional methods, and it has some application value.