Research On Image Mosaic Algorithm Based On Mutual Information

Along with the demand of modern medical science and the demand of the ultra precision machining and so on, the requirement of modern testing technology, especially the precision measurement technology is increasing day by day.Therefore To this end a series of testing instruments were invented, such as scanning electron microscopy, scanning tunneling microscopy, atomic force microscopy, confocal microscopy, and so on. The optical confocal microscope has the merits of non-contact, large field of view and high speed so it become the most influence measuring instruments,and it has been widely used in life science, ultra precision measurement, material science, medical science and other fields, to realize the measurement of material surface morphology, micro electric line width measurement, cell observation and other non-contact precision measurement. But high resolution is often relative to the large field of view, the large field of view of optical confocal microscope is also a micron level. And with the improvement of the resolution requirements, the field of view is reduced to nanoscale, so it is very easy to cause the object information cannot be observed in a view. Therefore we need to use image stitching technology to make the objects displayed in a picture in order to analysis the overall characteristics, and the confocal microscopy is ultra precision measuring instruments, its resolution is nanometer level, so the stitching precision requirement is very high and need to reach the sub-pixel level.In this thesis image stitching algorithm based on mutual information, aiming at the above problems research on image mosaic algorithm based on mutual information, is to study the mutual information measure function and optimization algorithm, the purpose is to improve the existing mosaicing algorithms to improve the precision, speed and accuracy, the algorithm is easy to fall into local extremum lead to false splicing, biological confocal microscopy confocal scanning high resolution two-dimensional image stitching measurement function, confocal microscopy confocal scanning high resolution 2D / 3D image mosaic, mosaic color pictures taken by CCD and 3D color image and color height CCD shooting wide field image stitching. The main contents are as follows:(1) In this paper, we propose three improved measure functions. The first is the direct improvement of the traditional mutual information measure function, increasing the edge information based on the analysis of its shortcomings, and construct the new measure function, new measure function curve contrast ratio increases, the higher accuracy than the traditional mutual information found in the registration experiment; The other two are indirectly through improved mutual information measure function the improved interpolation algorithm, improved first interpolation kernel function makes the local extremum curves obtained very good suppression measure, curve sharpness increase, it was found that the method can improve the accuracy of image registration results, and the suppression of local extremum makes the possibility to get the best transformation parameters increase; the third is to expand the field of interpolation algorithm and build a new interpolation kernel function, measurement function curve obtained by this method is smooth but the optimal transform curve sharp enough, through the experiment of the method The results obtained are larger, but the success rate is improved.(2) In this paper, I put forward a denoising algorithm based on statistical rule, using the proposed algorithm can effectively remove the shot noise of the image, and can be used with other filtering algorithm. I put forward an improved Retinex algorithm, the algorithm can effectively solve the traditional Retinex algorithm makes blurred edges of the image and the overall 'whitened' shortcomings, and by adjusting the contrast of the image obtained more bright colors?(3) Analysis of image stitching optimization algorithm using optimization algorithm PSO algorithm and Powell algorithm combines image mosaic optimal spatial transformation parameters solved and proposed for the shortcomings of the traditional PSO algorithm parallel PSO algorithm cooperation, experiments show that the algorithm speed and the success rate has improved greatly than traditional PSO algorithm, the standard function test can be obtained four times faster algorithm, the accuracy rate up to 80%.(4) By splicing a standard sample to prove the proposed method of registration sub-pixel precision level and the success rate can be guaranteed.