Research On UAV Orthophoto Maps Mosaic Algorithm Based On Graph Cut

Orthophoto Maps Mosaicking is an indispensable step in photogrammetric production work,the quality of mosaicking directly affects the extraction and analysis of image information.For UAV images,there are two main problems:the radiation difference due to the capturing angles,the lighting conditions,the sensor performance,and the geometric difference due to the projection difference.These two problems results cause the obvious brightness difference and geometric cracks in the mosaicked image.At present,the commonly used algorithms are difficult to handle these two problems at the same time,and some existing commercial software solutions is first generating initial seamline by mosaic algorithm,then human edit to obtain the orthophoto production meet the requirements.Which is not conducive to the automated production of orthophoto products.Taking the above two problems as the starting point and combining the characteristics of UAV orthophotos,the paper deeply studies the UAV orthophoto mosaic algorithm based on graph cut.The following is the main research work of the paper:(1)analyzed the reasons for the deformation of UAV images and ortho rectification methods,which was mainly analyzed the difference between the satellite platform and the UAV platform due to the height of the platform;analyze the reasons for the color difference in the image,and discuss the color consistency processing algorithm,including the processing of single image and between images;(2)researched on the construction method of d irected graph in image mosaicking based on graph cut,transformed the problem of image mosaicking into graph segmentation problem,and solved it with the help of graph theory.The main idea is to map the image difference matrix into a directed graph.The process of dividing the graph once by the graph cut method also completes the redistribution of the image overlapping area;(3)the weighting method of edges in the directed graph is studied,that is,the difference measurement method of the images in the overlapping area was studied.The directed edges were weighted by measuring the brightness difference and gradient difference of the pixels in the overlapping area and the local texture difference,difference calculation in multiple dimensions can better reflect the difference between images;This paper is focused on the orthophoto maps mosaicking algorithm based on graph cut method,which converts the image mosaicking problem into a graph optimization problem.With the help of the graph cut method of network flow in graph theory,get the global optimal solution also get the best seamline network.In the construction process of the map,a measurement based on the combination of the three aspects of the image color difference,gradient difference and texture complexity difference in the overlapping area is proposed,which can better reflect the image difference.Experimental analysis verifies the authenticity and effectiveness of the method,the mosaic result is better than the result obtained by dynamic programming,which can guide the seamline to avoid areas with obvious radiation differences and geometric differences between images,and obtain orthophoto products that meet the production requirements.