Research On Seam Line Detection And Quality Evaluation Methods For Orthophoto Mosaicking Based On Disparity Map

As one of the most important mapping products, digital orthophoto maps (DOM) can objectively show the details of the landform and the geographical objects while maintaining the correct planimetric positions. It is valuable for the geoscience and human uses, such as land resources survey, urban planning, infrastructure construction, earthquake relief work, etc. The traditional DOM production process includes single image dodging, digital differential rectification, image dodging between images, image mosaicking by seam lines and mapsheet division. The automatic production of DOM has always been the final goal for the photogrammetry field. A major bottleneck problem that hinders the automation process currently is the automatic detection and the network generation of the seam lines.A digital orthophoto map is generated by mosaicking a few digital differential rectified images, and in this process the seam line searching is the most critical step. The seam line should avoid crossing the buildings or the areas with color differences in order to make sure there is no geometry misplacement and hue difference on the result image, especially when dealing with large-scale urban images. Most of the traditional mosaic algorithms only consider the seam line searching within the overlap region of two adjacent images, and pay no attention on connecting the seam lines to form a network. Although some commercial softwares are able to create seam line network, but the kernel algorithms are not open source. At present, the seam lines are often manually edited after being generated by commercial software, but this operation mode results in low productivity so that it is facing a serious challenge in the currently rapid expansion of data. Therefore, it is so necessary to find out an efficient way to generate the seam line network and avoid crossing the building at the same time in the face of such a huge amount of data. In this paper, the automatic searching of the seam lines and the network generation are studied for the purpose of optimizing the production process.The mosaic quality is usually checked after the mosaicking is completed to ensure that there is no obvious geometry displacement or color difference. The traditional manual visual judgment is far too inefficient, so the objective quality evaluation methods are also studied in this paper.The main research work and innovative achievements are as follows:1) The generation and optimization algorithms of the disparity map are studied. The traditional seam line searching algorithms are mostly based on the differential image, which can substantially express the difference of the overlap region. But the description of the differences are not continuous, this would make the seam line pass through the buildings. This paper proposes to use the disparity map that generated by dense image matching to mark out the building area, and then marks out the timbered areas in the use of Moran’s Index based on spatial auto correlation. This method has achieved a good extraction result. The biggest difference between disparity map and differential image lies in the consecution of the areas, because the continuous regions accurately expresses the building areas and could be better used in seam line searching.2) Study on the intelligent algorithm for seam line automatic searching. The white pixels on the disparity map have high costs, represent the obstacle areas, while the black pixels represent the flat ground area with low costs. As a consequence, the searching process is to seek a shortest path on the disparity map giving a starting point and an ending one. In this paper, an intelligent algorithm based on greedy snake is used to searching for the seam line. Giving a starting point and a default step, the algorithm always find out the nearest point to the ending point at the circle that defined by the current point and the searching step. At last, orderly record the points to make up an optimal path. This method is different from the traditional way that based on comparing the pixel value in the four-neighborhood or eight-neighborhood. It is more efficient and can make sure the seam line pass through the buildings well.3) Study on the rapid generation and optimization algorithm of the seam line network. In general, the traditional pairwise mosaicking approach doesn’t fit the huge amount of data. This paper improves the network generating method that based on a principle that every pixel belongs to the image that with the nearest principle point. This method aims at the frame camera, and the basic idea is that the pixels have much smaller geometric distortion and much higher image quality when it is near to the principle point. The method divides the surveying area into many grids, and then computes the distances to all the image centers of every grid, the image index corresponds to the smallest distance will be the index of the grid too. Lastly, the grid indexes are aggregated and the boundary of the different aggregation planes is just the seam line network. Meanwhile, the original network should be optimized by the algorithm that based on the disparity map to avoid crossing buildings and trees.4) Study on the objective evaluation methods of the mosaic image. The evaluation of the mosaic quality is an important measurement of judging whether the mosaic result is equal to the subjective assessment by humans. The lack of the quality evaluation methods also limits the developing of the image mosaicking technology. The traditional statistical methods only work in ideal cases, therefore, this paper proposes an objective quality evaluation method based on edge detection, and it can better meet the subjective standards.