Research On Fast Detection Of Seamlines In Orthophoto Mosaicking

Digital Orthophoto Map (DOM) possesses real geometric coordinates and bears the features of high accuracy, rich information, and visual verisimilitude. Thus DOM is able to vividly reflect the topography, and becomes valuable product of geographic information. Therefore, DOM is widely used in many aspects including infrastructure construction, environmental monitoring, disaster relief work and so on. Automatic generation of DOM products is a tendency of photogrammetry and remote sensing. For the reason that simple orthophoto is usually unable to cover the area we interested in, orthophoto mosaicking become an inevitable procedure of remote sensing image process.To mosaick orthophotos, it is necessary to figure out the valid area each orthophoto contribute to the finally mosaicking DOM. And it should be guaranteed that there is no areas with multiple valid pixels or without valid pixel at all. That is, each pixel in the mosaicking orthophoto should be find in only one original orthophoto. The seamline network considers on the whole all the existing original orthophotos, and is order independent, does not create intermediate results. Thus, it is efficient and widely used.Usually the initial seamline network generated based on geometric information does not take the image texture into consideration. So it is inescapable that the seamlines may cross buildings, trees or areas with obvious color contrast, which causes fractures and displacements of buildings or unnatural color transition, and seriously affects the quality of DOM products. Manually modifying of the seamlines is a heavy task and will seriously restrict the production efficiency of DOM.For the problems in the mosaicking process of orthophotos, this dissertation aims to solve the following issues:1) Method to solve generation of the seamline network is researched as the valid area of the orthophoto being irregular polygon, or with complicate overlap, random arrangement. The method can be applied to orthophotos with large and small overlap. The orthophotos can be ortho-rectified image acquired from sensors on unmanned aerial vehicle (UAV), conventional aerial photographic plane, or satellite. The result can be used as the initial value of smart seamlines detecting, and also used to fast mosaicking of DOM, which caters to the demand of emergency surveying and mapping.2) Three ways are employed to detect the smart seamlines, avoiding crossing ground objects such as buildings, trees etc or areas of obvious color difference. They are:(1) Snake model, also called active contour model, is studied to search optimal seamlines. The seamline network is used as initial value. During the search procedure, Bresemham algorithm is used to calculated the energy of all the pixels the active contour passes. In this way, the method efficiently produces seamlines with high quality.(2) A method using Minimum Spanning Tree(MST) to search optimal seamlines is proposed. Calculate differential image which is deemed as a weighted undirected graph, according to the color intensity and gradient of the overlapping image. Using Prim algorithm to find the MST of the graph. When the ending pixel is add to the MST, an inverse traversal is adopted to extract the optimal seamline. In this way, the conventional iterative procedure, which faces high computational burden, is abandoned. Thus a fast and automatic method is achieved.(3) Watershed segmentation is studied to search the seamlines. A segmentation method in digital image process is introduce into orthophoto mosaicking. In the beginning, a mask is created for each orthophoto to mark its valid area. Then morphological image process is applied to the overlap image. After that, the watershed segmentation is carried out based on the corresponding masks, and the masks are updated afterwards. When all the orthophotos with overlaps are processed, mosaicking the finally orthophoto according to the masks.The advantages and disadvantages of the three methods are analyzed through experiments. It is proved that the proposed method overwhelms the other methods in quality of the seamlines and efficiency of the search procedure, and will effectively solve the problem of seamline selection in orthophoto mosaicking.