Study On The Sub-pixel Localization Of The Mark Point Based On Mixed Pixel Decomposition

The localization of target in digital photogrammetric images is not only an important research content of single image measurement but also a basic task of digital photogrammetry.The targets in photogrammetric images include both natural surface features which have obvious characteristics and mark points.The locating results impact subsequent work in photogrammetric image processing directly.The common mark points are studied mainly based on mixed pixel decomposition and are processed in matlab in this paper.The positioning accuracy of mark points can achieve sub-pixel by means of determining distribution inside mixed pixel depending on gradient information.The contents and achievements of this thesis are listed as following:(1) Sub-pixel edge extraction based on pixel decomposition.First of all,the relationship between target localization and edge extraction is stated,and several pixel level edge extraction operators which are introducted in the paper are experimentalized with images.The best edge extraction operator is chosed by comparing the experimental results.Then,mixed pixels on the edge of marks extracted from the digital image are decomposed based on the principle of mixed pixel decomposition by the linear model.The distribution inside mixed pixel can determined on the basis of gradient direction,which can improve positioning accuracy of the mark edge to sub-pixel level.(2) The sub-pixel localization of the circle target.The circle target is a common mark point.The first step is to extract pixels of the mark edge by the traditional edge extraction operator; secondly,these pixels of the edge are decomposed to sub-pixel points by the mixed pixel decomposition; finally,the center of the target are calculated through ellipse fitting based on least square with these sub-pixel points.This method has been tested by three experiments and the results show that its anti-noise ability is stronger and its positioning accuracy is higher than the traditional method that localization depending on pixel edge.(3) The sub-pixel localization of the corner.The corner is also a common mark point.The square vertical angle mark is studied mainly in this paper.Firstly,the corner is extracted by Harris operator to confirm its region,that is to say,to locate the square vertical angle mark roughly; Secondly,the mixed pixels are decomposed to sub-pixel points which can fit a straight line by Hough transform; Lastly,the intersecting point of two straight lines is calculated to be the sub-pixel location of the square vertical angle mark.In order to verify check the accuracy and superiority of this method,a method of sub-pixel edge extraction--gradient magnitude is introduced to contrast and analyse,three kinds of target images are experimented and the results prove that the sub-pixel localization based on mixed pixel decomposition is superior to the traditional method that locating target with pixel level edge,reach to the accuracy of gradient magnitude and more simpler than gradient magnitude.