Research On Pixel Level Image Fusion Based On Frequency Quantitative Extraction And The Assessment Methods Of Recognition Ability

Fusion of multi-resolution remote sensing images is needed to obtain remote sensing images with high recognition ability. The theoretical basis of image fusion to improve the recognition capability of remote sensing image is the combination of high frequency information and low resolution image, and accurate extraction of high frequency information and finding suitable fusion method are the key and difficult points to obtain high quality fusion image.In this paper, the image fusion is studied in frequency domain at the beginning. The research aims at the problem that the high frequency information is not given quantitatively in the existing fusion methods.The influence of image fusion in one-dimensional in frequency domain is analyzed firstly,then the research is extended to two-dimensional images, and the range of high frequency information in image fusion is quantitatively pointed out. Then two methods of image fusion based on frequency is proposed.A linear model is proposed to describe image fusion and be used in image fusion.In view of the shortcomings of the existing assessment methods, the paper proposes recognition ability as the evaluation index of the quality of satellite remote sensing images.The influence of image fusion on positioning accuracy is studied at last.The main contributions of this dissertation are summarized as follows:1.On the basis of analysis of various image fusion methods, the high frequency information of high resolution image is analyzed quantitatively, and the effect of the selection on the image resolution of the fusion image is pointed out. The research shows that under the same condition of methods and objects,the quality of fused images will be much better if the differences between high resolution images and low resolution images are much larger when the low resolution images are the same.For the same high resolution image, when the resolution of high resolution images are two times larger than the resolution of low resolution images,the image quality of fused images is closer to the high resolution images.2.Two image fusion methods based on frequency domain are proposed:image fusion method based on frequency priority and Image fusion method based on edge extraction.The basic principle of image fusion method based on frequency priority is to select the appropriate frequency and low frequency information. Then, the high frequency coefficients and low resolution image can be obtained by using the high frequency information. Image fusion method based on edge extraction is further extracted from the high frequency information, and the effect of high frequency information on the low frequency part of the low resolution image is reduced. Experimental results show that the method proposed in this paper is feasible.3.On the analysis of common fusion methods,a linear model is proposed to describe the process of image fusion and be used in image fusion.According to the need of people, two kinds of equations are used to solve the coefficients in the linear model.The experimental results show that image fusion method based on linear model is feasible, and the fusion results superior to conventional fusion methods.4.The quality of fusion image is evaluated by using the recognition ability. Based on the analysis of the existing evaluation methods, the paper proposes the method of recognition ability to assess the images fused by the optical satellite remote sensing images, recognition ability is divided into two aspects:the interpretation degree and the degree of recognition. The interpretation degree can identify the types of objects by GSD and NIIRS.The degree of recognition can judge the degree of difficulty in identifying the objects by image quality and edge clarity.Finally, the evaluation results are obtained by combining the two methods, and the feasibility of the proposed method is verified by using remote sensing images.5.The effect of image fusion on positioning accuracy is researched. Remote sensing image recognition ability and positioning accuracy are important indicators of the quality of satellite remote sensing image.There is a correlation between the recognition ability and the position accuracy.Image fusion can affect the positioning accuracy through recognition ability.Research shows that image fusion can affect the positioning accuracy, but it is not the main influencing factor.