Pixel-level Multi-focus Image Fusion Algorithm

Image fusion is a technique that integrates and processes several images from different sensors by some algorithm to obtain a new image which meets the requirements of some demand. The purpose of image fusion is to extract and synthesize images, so that more accurate, complete and reliable description of the images about the same scene and the same object can be acquired. The fusion image compensates the deficiencies of one isolated source image depiction, thus it is more suitable for human visual perception and computer processing.Multi-focus image fusion is a technique that obtains a completely clear image, which is of the same image-forming condition and the same scene but with different focus points, by the multi-focus image fusion algorithm. Nowadays, there are a lot of multi-focus image fusion methods, mainly consisted of the conventional fusion method, multi-resolution analysis method and intelligent method.This paper concentrates on multi-focus image fusion algorithm on pixel level, and offers penetrating study and exploration of image block method, wavelet transform method and curvelet transform method. This paper introduces the image block segmentation into the multi-focus image fusion, and uses wavelet spatial frequency as the multi-focus image clarity indicator. We can determine the clarity of the image by calculating the wavelet spatial frequency. Therefore, the image is divided into a clear region, a fuzzy region and the junction of the two regions. In the multi-focus image fusion method that is based on image block, the pixel of the source clear image block will be directly applied to the pixel of the fusion image. As to the selection of the pixel value of the junction of the two regions, this paper altogether proposed two algorithms.In the first algorithm, the spatial frequency window is calculated and the larger pixel value is used as the pixel value at the junction. In the second algorithm, the pixel value corresponding to initial fusion image evolved from wavelet transform is directly chosen as the pixel value at the junction. The experimental results show that, these two algorithms improve the fusion quality and effect.Curvelet transform, as a new multi-resolution analyzing method, inherits the fine local features of spatial and frequency domains of wavelet transform, and overcomes the defects of wavelet transform in depiction of the direction features of image edge. We analyzed the first and second generation curvelet transform, and described the process of the second generation curvelet transform. In order to better illustrate the superiority of curvelet in the description of the image edge, we compare the fusion results based on wavelet transform and curvelet transform under the same fusion rules. The experimental results show the fusion image based on curvelet transform is clearer and has a smoother edge. We proposed two improved algorithms to improve the effects of curvelet transform.The first one is to combine the curvelet transform with the image block segmentation, and to divide image into three regions, namely, a clear region, a fuzzy region and the junction of the two. The pixel of the source clear image block will be directly applied to the pixel of the fusion image. As to the junction, the pixel value corresponding to initial fusion image that is evolved from curvelet transform is directly chosen. The experimental results reveal that this algorithm can get much better fusion results than that by using curvelet.The second is to combine the curvelet transform with the regional energy. Firstly, two source images are decomposed by using curvelet transform. Then, in choosing the low frequency coefficient, fusion regular of the weight averaging is employed. While, in choosing the high frequency coefficient, fusion regular of the maximum of the regional energy value is employed. Finally, the fusion coefficients are reconstructed to obtain fusion image. This algorithm takes the neighborhood correlation between pixels into consideration, therefore, it can improve the fusion effects.The last is the summary of this research and the prospects of future work.