Key Technology On Color Balancing For Creation Of Color Consistency Synthetic Products With Optical Remote Sensing Imagery

With the increasing number of remote sensing satellite observing platforms,more and more satellite imageries are obtained.In the application of earth observation,the satellite imagery covering large areas have become increasingly important.Due to the limitations of sensor design,the coverage of a single image is limited.To obtain images with large coverage,image mosaicking is usually adopted using images obtained at different time and even from different sensors.Due to several factors,such as solar incident angle,atmosphere,and illumination condition,the radiances between images are different,thereby posing a large problem for the creation of seamless mosaic imagery.In order to eliminate the differences between images,the process of color balancing method is necessary.There exist numerous researches about color balancing,however,some problems still exist in these methods.These existing color balancing methods can be categorized as absolute radiometric calibration and relative radiometric normalization.Both absolute and relative radiometric calibration methods have advantages and limitations.The absolute radiometric calibration is able to convert the digital number value of the image to the accurate surface reflectance.However,the rigorous conditions required for absolute calibration are usually difficult to obtain,thus making absolute radiometric calibration of a great number of images impractical,and narrowing down its application range.Despite the advantage of its simplicity and practicability,relative radiometric normalization methods have disadvantages as well,such as the problems of choosing suitable reference image,color error propagation,and the two-body problem.Moreover,the color balancing process is one of the step of the creation of seamless mosaic imagery,and the mosaic images has different applications.Thus,it is necessary to choose the suitable process method for different purposes,which is neglected in current researches.Under this condition,three color balancing methods are proposed in this paper according to the situations of image data and its application for the purpose of eliminating the color differences among different images.The main research contents are as follows:(1)A mixed radiometric normalization(MRN)color balancing method is proposed in this paper.The proposed method combining the absolute radiometric calibration and relative radiometric normalization is achieved by the following steps.First,the middle-resolution image after absolute radiometric calibration is selected as the reference image according to the acquisition time and the geographical coverage of the target image dataset.Second,the invariant probability of both the target image and the reference image is extracted using the approach of the linear invariant IR-MAD(Iteratively Reweighted Multivariate Alteration)method.Third,the target image is separated into image tiles by certain rules,which then enables building a relative radiation regression model by pixel points with corresponding invariant probability in each image block.Finally,the radiometric correction parameters of each pixel are obtained by bilinear interpolation,from which the calibrated images can be generated.The novel MRN approach proposed in this paper combines the precision of absolute radiometric calibration and the adaptability of relative radiometric normalization,while effectively avoiding the limitations of absolute radiometric calibration as well as the uncertainty and the two-body problem of relative radiometric calibration.Moreover,to a certain degree,choosing the surface reflectance product as the reference to correct other images reduces the influences caused by the atmosphere and other factors which makes up for the limitation of absolute calibration and is beneficial to the remote sensing quantitative analysis and research.(2)A novel auto-adapting global-to-local color balancing method(AGLCB)is proposed in this paper.The proposed AGLCB method combines both the global and local optimization strategies to eliminate the problem of color difference existing in different images in the aspects of the elimination of overall color difference and the optimization of the local color detail information.The global optimization strategy takes the assumption that the color information in the overlap area of the adjacent images should be equal after the color balancing process,and takes the constraint that the color information of the image before and after the color balancing process should be minimal.The global optimization strategy takes all target images as a whole and solves the normalization regression models simultaneously without the requirement of reference images,which transfers the problem of the elimination of color difference into the problem of least square optimization problem and can eliminate the total color differences effectively.However,due to the neglect of the local information of the target images,residual color differences may still exist in the overlap areas of the global optimization results.The local optimization strategy is a supplement for the global one,which focuses on the local information to eliminate the color differences in the overlap areas of the target images with the Gamma transform algorithm.The proposed AGLCB method can effectively eliminate the problem of color difference between images.The color information of the color balancing results is consistent and smooth,which is approximate to the color spectrum characteristics of the original images at the same time.The proposed AGLCB method is more suitable for the creation of seamless mosaic image for the purpose of the remote sensing visual interpretation issues.(3)The establishment of color reference library and the color balancing method based on the library are proposed in this paper.A color reference library is a database where color and other related information of the existing images is organized and stored under certain rules.The images which are selected to store in the color reference library are the existing mosaic datasets or other images with the color information which is visually appealing and consistent with the human visual perception.The data source is organized according to the factors of geographical coverage,acquisition time,and spatial resolution.Since it is the color information rather than the image itself is stored in the color reference library,the storage of memory is greatly reduced and the library is portable.The proposed color balancing method,making full use of advantages of the colour reference library,is able to select reference colour data with the same geographic coverage,acquisition season and the closest spatial resolution automatically and process the target image adaptively.The strategy of colour segmentation is applied in the proposed method,which is able to reflect the colour distribution of the images and performs better than the color information extracted from the image as a whole.In order to get a smooth color balancing result,each pair of linear coefficients is weighted by a color influence mask.The colour reference library makes up the blank of the research about the color information organization,by which the problem of choosing reference images in traditional methods can be solved.Since the colour of the images corrected by the proposed method is very close to that of the reference data which is consistent with human vision habit,extra work of human interaction on colour is not required,which is time-saving.In order to verify the correctness and reliability of the proposed methods,multiple experimental datasets are used to test the key points of the proposed three color balancing algorithms.Experimental results show that the proposed three color balancing methods can deal with different application scenarios,and can effectively eliminate the color differences between different images,which lay the foundation of the creation of color consistency synthetic products and its subsequent applications.