Multi-focus Image Fusion Method Research Based On Multi-resolution Singular Value Decomposition

Image fusion is an advanced image processing technology, which integrates the information generated by multiple sensor about the same scene and objects, or by the same sensor in different time and creates a new image according to some rules. Different images have different spatial resolution, spectral resolution. Image fusion technology not only retains the important characteristics of the source images, can also process and use the multi-source information cooperatively to enables the different form information complementary, thus improve the ability of analyzing and extracting the image information, and recognize the same thing or object more objectively and more constitutionally. With the rapid development of computer information process technology, the application of image fusion technology is becoming wider and wider. It can be used in the fields of military affairs, medical science, remote sense, machine vision, etc.At present, the multi-resolution analysis method such as wavelet transform is widely used in image fusion. But wavelet transform cannot make full use of data unique features itself, sometimes can only catch limited direction information, especially the tensor order wavelet, which cannot make full use of its image geometry regularity since the lack of direction. Although non-separable wavelet can realize isotropic, but there is still a defect of large computation. In recent years, singular value decomposition attracts more and more scholars to study for its optimal de-correlation and low-rank approximation characteristics. Multi-resolution singular value decomposition (MSVD) is a new multi-scale transform. It has good multi-resolution characteristics and the linear computation complexity. To this end, this paper presents a new framework of multi-resolution singular value decomposition. And it is applied into multi-focus image fusion. The multi-focus image fusion based on the four channels MSVD is proposed. Firstly, according to the principle of blocking algorithm, images involved in fusion are decomposed into one approximation and three detail images with different resolution by singular value decomposition. Then combined with reconstruction algorithm of MSVD, the frame of image fusion is given. Secondly, compared with image fusion by discrete wavelet transform (DWT), image fusion by MSVD performs better. Moreover, MSVD does not have a fixed set of basis like wavelet, and its basis vectors depend on the image itself. At last, the performance of the result image was evaluated using objective indices. The experimental result shows that the proposed method is not only simple, but also that the visual effect of the image is considerable after reducing blocking artifact. And both the definition and spatial frequency are improved greatly.In order to reduce the blocking artifacts in the fused image with MSVD, a block ing-artifact reduction algorithm based on mathematical morphology edge detection is presented in this paper. Firstly both edges of the two original images are figured out using morphological gradient operator. The image fusion is performed for the two edges. Then the fused edge image can be as the edge of image with blocking artifacts. Secondly the final result image with no blocking artifacts is obtained using inverse transformation of morphological gradient. At last, the performance of the result image evaluated using objective indices. The experimental result shows that the proposed method is not only simple, but also uniforms the image. After reducing blocking-artifact, the visual effect of the image is considerable. It also shows that both the definition and spatial frequency are improved greatly, and is more powerful in edge-preserving.The paper extends the four channels MSVD method, and presents the six channels and the eight channels method of MSVD. Both methods are applied respectively in the multi-focused image fusion. The image is decomposed in a finer and wider frequency range, which makes the fusion result image to be more fully integrated source image information. The two kinds of methods decomposed the source image into a low frequency sub-image and several high frequency sub images according the blocking algorithm. The six channels MSVD gets five details using the block of two by three. And the eight channels MSVD gets seven details using the block of two by four. In the fusion process the low frequency images fusion use the simple weighted average method, and the fusion rules of the high frequency images is selecting the larger absolute value. The experimental results show that the fusion effects of the both methods are better than that of four channels, further more the resolution and spatial resolution are obviously improved, and can keep the image structure information better.