Research On Optimal Color Space Selection Of Buildings Based On Bayer True Color

With the rapid development of aerial photogrammetry,images acquired by various sensors equipped on lightweight and unmanned aerial platforms become a powerful supplement of satellite remote sensing images.Bayer true color image is one of the main types of such images.In general,there are only three bands in Bayer true color images which have the advantage of high resolution and the shortcoming of poor spectral information.The traditional pixel-based extraction method may face challenges when deal with such images.The object-oriented approach which can take advantage of image texture and shape information may have the ability of address this challenge.On the basis of homogenous objects obtained by multi-resolution segmentation method,the increasement of spectral information utility and the enhancement of target information according to proper spatial transform method would greatly improve the extraction accuracy of target information,which would be further studied.This article,on the basis of previous studies,attempts to process the interpolated Bayer images with linear transform space(YIQ,YCbCr,I1I2I3)and nonlinear transform space(HSI,CIE(L*a*b*),Nrgb).With the visually interpretation of the results of six color space transformations,it showed that the YIQ,I1I2I3,CIE(L*a*b*)color space can achieve better segmentation results.Of the computed correlation of three bands in each transformed image,the I1I2I3 is the smallest.According to the principles of greater correlation had greater impact on classification accuracy,I1I2I3 is initially supposed as the optimal color space for segmentation.The paper firstly took five types of non-collapsed buildings as test data,including cottage,tile-roofed house,building,tent and plant.The transformed images of six color spaces were segmented at a series of scales,and then evaluated by qualitative and quantitative methods.Qualitative evaluation conducted a comparison of the difference of the same building type in different color spaces and scales,with better segmentation results in YIQ?YCbCr?I1I2I3?Nrgb color space.Quantitative assessment took factors such as the characteristics of RGB images and the outline shape of the building into account,and built three objective functions corresponding to topology,geometry and spectral area.For the slightly difference of optimal segmentation space of different types of non-collapsed buildings,a multi-objective constraint function was introduced to comprehensive analyze all buildings.It came to a conclusion that the I1I2I3 space was the optimal color space for segmentation.In order to verify the correctness of this conclusion,the paper selected 10 true RGB Bayer color images to find out the universality of the result.In addition,with the consideration of the important role of UAV played in the earthquake relief,the paper carried out a series of processing with the collapsed houses after the earthquake on the images.The type of buildings was not considered and calculated anymore due to the irregular fragments of building objects.The corresponding experimental showed that six color space transform had little impact on the classification accuracy of collapsed buildings.On the basis of the evaluation of the optimal color space for non-collapsed and collapsed buildings,the paper conducted a classification with images after the earthquake.Beyond the original RGB image,the I1I2I3 transformed image was also segmented and classified by thinking over the geometry shape and texture characteristic parameters.The comparative analysis of the extraction accuracy of the color space transformed image and the original image showed that both the overall accuracy and Kappa coefficient of the latter can reach a value of more than 0.8,as well as the two indicators of I1I2I3 transformed image can be more than 0.95.It can be concluded that the proper color space transformation can improve the extraction accuracy of useful information to a great extent,which would indirectly determine the wider significance and effectiveness of the optimal color space.