Research On Image Mosaic Technology Based On Point Features

With the development of science and technology,people are so eager to get higher quality information to enjoy a better life.In the sense of human,vision is the most important way to obtain information.And image is the main way of carrying visual information.It's very difficult to get large view images because of the weather conditions,camera performance,shooting height and other complex factors.Therefore,in order to obtain high-quality visual experience,image mosaicking technology emerged.Image mosaicking technology is to mosaic two or more images with certain overlapping areas to obtain accurate an image with wide filed of view.It is an important research topic in the fields of image processing,pattern recognition,computer vision and computer graphics to create favorable conditions for geodetic survey and geographic survey.Image mosaicking technology can be classified according to different standards,among which feature-based image mosaicking technology is the most concerned.And in feature-based image mosaicking technology,point-based image mosaicking technology has become a hot spot in image mosaicking technology.The extraction and description of feature points are the most important basic steps in the whole process of image mosaicking technology.The innovation of this paper is mainly focused on the extraction and description of feature points.The classical algorithm is improved and combined.All the innovation is used to realize image mosaicking,and the results are evaluated.The specific work is as follows:(1)The principle of image mosaicking is studied.The whole process of image mosaicking includes image preprocessing,image registration,image fusion and image quality evaluation.In image preprocessing,image denoising,image enhancement and geometric correction are mainly introduced.In image registration,image transformation models and the methods how to get them are introduced.In image fusion,the chroma adjustment method based on overlapping areas and common image fusion methods are introduced.In image mosaic quality evaluation,PSNR(Peak Signal to Noise Ratio)method,SSIM(Structural Similarity Index)method and DoEM(Difference of Edge Map)method are introduced.(2)An improved fast corner detection method was proposedOn the premise of judging whether a candidate is a feature point according to FAST(Features from Accelerated Segment Test)method,the law of gravitation is applied to digital image processing to determine the direction of the resultant force of the candidate point.And the direction can be introduced into the process of feature point detection to preliminarily judge the candidate point.And the,the speckle noise can be removed according to itself characteristics.Carried experiments by using SAR images to show that this method can suppress speckle noise in a way.(3)A feature point selection method combining SIFT and FAST algorithm.According to SIFT(Scale Invariant Feature Transform)algorithm to extract two sets of feature points of reference image and moving image,respectively.And then,using FAST algorithm to get another two sets of feature points of reference image and moving image,respectively.The common feature points on the same image are selected by the two methods.Imaging matching is conducted based on the common feature points.And the transformation matrix is calculated to transform the moving image.The moving image can be transfor according to the transformation matrix.Finally,the weighted fusion algorithm is used to complete mosaic mosaicking.This method not only retains the feature points with scale information detected by SIFT method,but also shortens the time of describing unuseful feature points.The experimental results show that the combination method has advantages in image matching and improves the quality of image mosaic.(4)An image mosaicking based on multi-feature descriptors is proposed.The feature descriptor based on vertical and horizontal gradient,which can capture the change of gray level in local area.The feature descriptor based on color invariant information,which have filtering function for the color image.The feature descriptor based on pixel intensity,which can be used to describe the dependence relationship neighboring pixels.Four feature descriptors have their own characteristics.Combining them can improve the uniqueness of feature descriptors.The proposed multi-feature descriptor combines local vertical and horizontal gradient,color color invariant information and pixel intensity.The experimental results show that the proposed method can improve the quality of image mosaicking.