Research Of Image Fusion Algorithm

The target of image fusion is to combine feature information from multiple images under the same scene and create a new image about this scene. The fused image provides observers with more reliable and complete information than that is offered by a single sensor. Just because of the characteristic of image fusion technology,it has been widely used in various fields, such as computer vision, feature extraction, segmentation, and medical image processing, etc. Image fusion can be performed at three different levels: pixel-level, feature-level and decision-level. In this paper, the pixel-level and feature-level fusion methods are mainly studied. Aiming at the problems in the existed methods, some new image fusion methods are proposed in this paper. The main work of this paper is as follow:①. As for the problem that the texture information can not be extracted effectively using Curvelet, a multi-focus image fusion method by combing the PDTDFB (Pyramidal Dual-Tree Directional Filter Bank, PDTDFB) and the wavelet is proposed in this paper. By using its multi-scale, multi-direction and shift-invariance features, the defects of Curve let can be eliminated.②.Aiming at the computational complexity, low efficiency in Non-subsampled Contourlet Transform and the aliasing problem in Contourlet domain, the Non-aliasing Contourlet Transform for image fusion using cycle spinning is proposed in this paper. The Non-aliasing Contourlet Transform is introduced into the transform domain for image fusion, meanwhile cycle spinning is applied to improve the translation invariance and overcome false information of fused image. The results show that the method is better to remove the image aliasing effect in Contourlet transform and reduce the time consuming. The visual effect has significantly improved.③.Allusion to the incomplete orientation information of support value transform and that the salient feature can't be obtained by the filter generated by Gaussian kernel. An improved method for the multi-focus image fusion which is based on the support value transform is proposed. It brings the PDTDFB into SVT transform domain to extract the direction feature of image, and at the same time, the Gaussian kernel is replaced with wavelet kernel. Experiments proved that orientation and edge features can be extracted effectively by using this method.The above three methods have advantages and disadvantages of their own, and they are applied in different applications. The algorithm based on the PDTDFB and wavelet has good performance in Objective evaluation, low time-consuming, so it is suitable for the real time system. The algorithm based on the Non-Aliasing Contourlet Transform for image fusion using cycle spinning performed best in visual effects; it's suitable for demanding visual system. The algorithm based on Support value transform is efficient to extract the salient features of images, and can improve the visual impression. But as it's rather complex and time-consuming, so it is suitable for system with high demand for visual effect and low demand for time-consuming.