SIFT Feature Based Extraction And Optimization Algorithm For Mosaicing Electron Microscope Images

As it is well known to us,the technology of computer graphics has been advanced rapidly,which makes people explore deeply in all variety of fields by developing new devices.For example,film and television animation industry hopes to make marvelous scenes by processing original graphics,virtual reality technology hopes to give users an immersed feelings and etc,which cannot go without image mosaic technology.If we take photos with all scenes in it,we will always get roughly image and cannot get delicate parts.So it's necessary for us to get high-precision images as much as possible by developing devices,which can be used to generate big-size and easy-to-observe images.Now the image mosaic technology is relatively mature and has been applied widely.The next step of this technology is more in the intersection of different subjects and fields,and the optimization of performance.The steps of Image mosaic are firstly extracting features by creating Gaussian scales pyramid,then matching features from different pictures;lastly merging the good matched images and getting the final picture.Based on the SIFT feature which is insensitive to the transformation of rotation,scale,zoom and luminance,and is also relatively steady to the transformation of viewpoint,affine and noise,last but not the least the extraction of features from image is rich and the speed of the extraction is faster than lots of other algorithms.Then we can use the RANSAC algorithm to find effective match by the features which found in last step.Finally stitching the images and getting a panorama.This work is devoted into putting an optimal algorithm based on the pictures which are from electron microscope.Because SIFT features is steady in a large range of viewpoint transform.So we can get a large of SIFT features from different images.Images from microscope must have the intersection which is used to match,and also have different part which is used to record more information.So we only need to scan the intersection and get match features which need help from the scanner's movement of electron microscope.Large experiments tell us that this algorithm can improve the speed of finding effective SIFT features,and the average rate is more than 35% and also reduce the pressure of the RANSAC algorithm.