Research On Algorithm Of Remote Sensing Image Mosaics

In order to expand field of vision and obtain higher quality and more precise position information in remote sensing image applications, two or many remote sensing images need be spliced one image. The remote sensing image mosaics have been commonly used in many fields, e.g. environmental monitoring, land resource investigation, vegetation resources investigation, land water resources investigation and marine resource investigation. Therefore, the remote sensing image mosaic technology has become one of the basic techniques of regional monitoring and macroscopic rule research using remote sensing in geography science. In this dissertation, two key techniques of remote sensing image mosaic, which are image registration and image fusion, have been focus on.The decision of remote sensing image registration will directly affect the validity of mosaic result, so image registration is the precondition and base of remote sensing image mosaic. In this dissertation, the performance comparison and analysis of three typical similarity measures for remote sensing image registration (sum of squared differences, correlation coefficient, and mutual information) have been done on the basis of computing time, sharpness curve and effect of image window size. The experiment results show that mutual information may reflect well the related information between the images, but it has two shortcomings. One shortcoming is that it is easy to appear partial extreme value when computing the maximization of mutual information, another is heavy computation load. Therefore, a novel method is proposed to solve these two shortcomings. Edge gradient information is extracted from remote sensing images based on GRA and integrated with mutual information to reduce the appearance of partial extreme value. An image pyramid is built by the wavelet transform, and then coarse-to-fine search strategy is used to reduce computation load. The experiments of the remote sensing image registration using our method are satisfactory.Edge feature is one kind of very important features in multi-source remote sensing image registration and is the foundation of image registration based on edges. An algorithm is proposed, which makes full use of tensor voting and multi-resolution algorithm to extract the large edges of remote sensing images, and combines the accurate localization of the tensor voting in high resolution and the anti-noise performance in low resolution after wavelet transformation to detect more complete edges. The proposed method has provided one kind of steady match characteristic for multi-source remote sensing image registration. The approximate alignment parameters of two images are found with partial hausdorff distance similary measure in coarse regisition process, and then the best alignment parameters are obtained in fine regisition process.In recent years, the optimized joint algorithm has been used for seamless texture mosaic. The main idea of this algorithm is finding a cut line in the overlap region which the texture and color error is least between the two input images. A good cut line may reduce or even avoid the ghost and fuzzy effect in the image mosaic caused by texture difference. The dissertation has deeply researched the optimal seam-line detection algorithm and has presented two improved methods.This dissertation puts forward a novel algorithm based on grey relational analysis for detecting an optimal seam line in remote sensing image mosaic. First, the algorithm regards pixel neighboring sequence in overlapping area as referential sequence and compared sequence because of the relation of center point and neighboring points, then computes grey relational degree of both sequences using slope relational degree, finally finds an optimal seam line which has less differences in color and texture between the images. The experimental study indicates that the method presented above is more superior to equal division method.The technology of detecting an optimal seam-line in overlapping region between the input images based on dynamic programming is less complexity than others and has simple idea. An improved method is putted forward. In order to consider the similarity of geometry structure more roundly, the gradient operator is replaced by grey slope relational degree and the new criterion computing easily and stably is obtained.Due to the color difference caused by some factors such as small changes in the environment and different point of view and exposal, the seam effect is obvious after stitching two remote sensing images. The advantages and disadvantages of two kinds of seam-line removal methods, one is based on wavelet transform and another is based on overlapping image, are analyzed. Then the hard correction method based on color spatial transform and dynamic width is used. In HIS color spatial, the value of pixels of seam-line in the whole stitching image is removed in hue, luminance and saturation, and then transformed to RGB spatial. The experimental results show that the method is very effective, simple and easy to be realized.Remote sensing image registration, seam-line detection and seam removal have been studied in this dissertation. New methods putted forward might be a help to research work and application of remote sensing image mosaic in future.