Image Fusion With Edge-preserving ?ltering

Image fusion is the technique that combines relative valid information of source images into a single synthesized image, thus making it a more complete image with valid information than any input source images. One of the common image fusion methods is the image fusion within transformed domain, which turns the source images into coe?cients of the transformed domain and fuses the corresponding transformed domain coe?cients of each source image before inverse transformation to reconstruct the composite image. The advantage of fusion in transformed domain is the good capture of details for different scales and thus makes it popular in application and development. But a common problem of transformed domain fusion is the intensity variance after fusion that prevents it from maintaining the original intensity of source images and even results in intensity out of original range or contrast distortion, which affects the presentation of salient information in serve case.With the property of bilateral ?lter and guided ?lter, 3 image fusion methods are proposed in spatial domain. The methods make use of the difference between the original image and its local average ?ltering, image gradient, and Laplacian energy as the main feature for saliency measure respectively, and the average?ltering or block-wise processing gives the saliency measure the ability of covering the main information area. Comparison between saliency measure of original images is used to obtain the initial primitive fusing weight, but the initial weight is too crude to do the job for its poor local selection, so the edge preserving ?lter combining the original images comes to overcome the weakness and align the weights with the salient edges of original images respectively. The fusion weight is re?ned and matched with the salient area by combining the maximum scheme and the conditional edge transferring property of edge-preserving ?lters, which can avoid the distortion of intensity and contrast. By ?ltering the fusion weight with edge-preserving ?lter, the problem for unnatural edge transition of adjacent areas and incomplete detail selection can be solved. Besides it’s more e?cient for fusion in spatial domain than in transformed domain.Experiments with several different fusion methods on multi-focus and medical image fusion validate the effectiveness and the e?ciency of the proposed fusion methods using edge-preserving techniques. The methods are compared and analyzed for their advantages and disadvantages, results show the advantages of the proposed methods for both time consuming and high quality.