Seamline Extraction For Remote Sensing Images By Incorporating SURF And Watershed Segmentation

High-resolution remote sensing images have greatly broadened the spatial scaleof earth observation and have been widely used in cadastral survey, landusemonitoring, urban planning, and disaster relief, and so on. However, with theimprovement of the spatial resolution, the image width is severely restricted, a singleimage is increasingly unable to meet the needs of various researches and applicationsof remote sensing technology. Therefore, the importance of automaticallyimagemosaic technology is becoming more and more obvious in various remotesensing applications.A seamlineextraction method mosaic for remote sensing image mosaic ispresented in this paper, and the main work includes:(1) The advantages and disadvantages of classic SURF in automatic matching ofhigh resolution image are analyzed. On this basis, a space constraint SURF featurepoints matching method is proposed, which improves both the speed and precision offeature points matching andmakes the feature points tend to be uniformlydistributed.The algorithm firstly applies dual matching process between the centralblock in the target image and the base image. Then, the initial geometrictransformation parameters are calculated. By limiting the search space of feature pointmatching, both the accuracy and speed are improved. Finally, structural similarity ofimages is used to identify tall objects, and points located on these objectsare thusdeleted to avoid those feature points misleading the calculation of the transformationmodel between images.(2) A more flexible coarse seamline generating methodfor multiple imageispresentedbased on region growing. The method is easy to implement, and have almostno limitation to the shape of effective overlap region, which can be simplequadrilateral, general polygon and more complex shapes.(3) Based on watershed segmentation, a seamline optimization method whichtake into account the similarity of adjacent segmentsis proposed. By makingtheseamline fall on edges ofsignificant featuressuch as waters, roads and buildings, thisalgorithm is able to maintain the integrity of surface features in the final mosaic imageand thus achieve seamless mosaic.Firstly, the.maker-controlled watershedsegmentation based on morphological reconstruction by opening and closing isapplied to sketch edges of ground surface objects.Then, take the three-way intersectionsand four-way intersectionsofthe watershedpath as shortest path searchnodes, which reduce the number of nodes by at least four orders of magnitude andthus greatly reduce the time and space complexity. Take Transformed Divergence andJeffreys-Matusita Distanceas patch similarity of adjacent segments, path betweenadjacent segments with high similarity will be punishedto ensure seam line will belocated atedges of significant features.In this paper, IRS, ALOS images (medium resolution) and UAV images (highresolution) were used to validate the proposed method. The results indicated that theproposed feature point matching algorithm is fastand robust,the extractedseamlinetend to follow edges of salient ground objects and the final mosaic imageshave good visual effect.