Study On Remote Sensing Image Fusion Based On Super-resolution Reconstruction

Remote sensing system generally carries two sensors, multispectral sensor and panchromatic sensor. The first one provides low spatial resolution of multispectral remote sensing images, and the second shows high resolution panchromatic images, while those do not contain multispectral information. In practical application, it seems not easy to achieve the desired effect by getting remote sensing information from only one single sensor.Along with the rapid development of sensor technology, the needs and eagerness of obtaining effective information are significantly increased. It could extract the correct information in high efficiency, but only if all sorts of multi-source date are fully utilized. Image fusion technology is an image processing techniques that compounds the multi-source images as a single, enhanced picture, which offers added value to achieve the better result.Through fusion of two kinds of image data, more information will be merged into the same data set. The main purpose of this study is not only to enhance the apparent information in the respective source images, but also to increase the reliability of target detection and automatic target recognition.In this paper, we took the QuickBird images as the research object, and studied the super-resolution reconstruction, then fused the high resolution panchromatic image with low resolution multispectral image, in order to get higher resolution of multispectral images. The major jobs of this paper are the listed as following:(1) This study investigates the symmetry properties of image degradation model which called point spread function (PSF). A simplified image degradation model is deduced. By using Fourier transform, the module of Point spread function (PSF) has the quality of symmetry. Based on the theory, we gave a simply calculating method for convolution computation in the process of degradation.(2) Because of the panchromatic image is modeled as a linear combination of the ideal multispectral images. The relationship between the panchromatic band and multispectral bands is considered to derive the estimation of point spread function.(3) In according to the image blind restoration theory, the paper puts forward the point spread operator matrix and the fusion estimation algorithm of the image and get the high resolution of the fused multispectral image. During the Image reconstruction, the point spread function and the fused image are both unknown. Using the characteristics of the discrete sine transform, the SPF and fused image could be joint estimated.(4) Using the K-VSD training redundant dictionary, the expression of the original image sparse, and through a small measure, to restore the ideal scenario high resolution multispectral image, and the color image and infrared image fusion, get better image visual effect.(5) The traditional compression perception algorithm is improved. The ideal high-resolution multispectral images are estimated based on the sensed multispectral and panchromatic images. The orthogonal matching pursuit algorithm is simplified.(6) By comparing with the traditional algorithm via simulation experiment in the MATLAB platform, the super-resolution reconstruction algorithm introduced in this paper not only improve the resolution effectively but also maintain the spectral information of the original image.