Effect of Ground Control points Location and Distribution on Geometric Correction Accuracy of Remote Sensing Satellite Images

Transcription

1 13 th International Conference on AEROSPACE SCIENCES & AVIATION TECHNOLOGY, ASAT- 13, May 26 28, 2009, Military Technical College, Kobry Elkobbah, Cairo, Egypt Tel : +(202) , Fax: +(202) Paper: ASAT-13-RS06 Effect of Ground Control points Location and Distribution on Geometric Correction Accuracy of Remote Sensing Satellite Images F. ELtohamy *, E. H. Hamza ** Abstract: Remote sensing imagery, from satellites, is inherently subjected to geometric distortions. Therefore geometric corrections, as preprocessing operations, are normally required prior to imagery analysis and extraction of information. The most common approach for geometric correction is the use of mapping polynomial.. It depends on selection of several clearly discernible points, called ground control points (GCPs), in the distorted image, and map them either to their true positions in ground coordinates (e.g. latitude, longitude) measured from a map, or to georeferenced image (corrected before), coordinates of corresponding points, through a mathematical transformation, that will convert the raw image coordinates into the desired coordinates. The accuracy of the geometric correction process depends mainly on the number of selected GCPs (polynomial order), the identified features on the image (road intersections, airport runway intersections, bends in rivers features and the like), and on the distribution of the selected GCPs over the distorted image area. The effect of number of GCPs on geometric correction accuracy was presented [1] by the authors and others. In this paper the effect of the selected location of GCPs (the identified features on the image: road intersections, airport runway intersections, bends in rivers features and the like or the others), and the way of distribution of the selected GCPs over the distorted image area on geometric correction accuracy are presented. The Root Mean Square Error (RMS), are calculated and used as a measure of accuracy of the obtained results. Keywords: Remote sensing satellite imagery, Geometric correction, Image registration, Ground control points, Resampling. 1- Introduction: All remote sensing images, from satellites, are subjected to geometric distortions. Therefore geometric corrections, as preprocessing operations, are normally required prior to imagery analysis and extraction of information. * Egyptian Armed Forces ** Egyptian Armed Forces, esamhamza@hotmail.com 1/14

2 Geometric distortion in an image means that the image features positions do not accurately relate to scene on ground or on map positions. Earth rotation during imaging acquisition, Earth curvature, and satellite altitude, attitude, and velocity variations are the main sources of geometric distortions [2]. Geometric distortions can be classified into two types [3], [4]: a. Systematic (predictable) distortions. b. Non-systematic (random / unpredictable) distortions Geometric corrections are intended to compensate for these distortions so that the geometric representation of the imagery will be as close as possible to the real world. There are two main approaches [3], [4], for correction of the various types of geometric distortions. a- Systematic distortions can be corrected by applying formulas derived by modeling the sources of the distortions mathematically and use these models to establish correction formulae. These modeling techniques require priori knowledge about the orbit parameters; the nature and the magnitude of the sources of distortion during the scene acquisition time. Sometimes this priori information is not available and consequently these techniques cannot be applied. b- Random distortions and residual unknown systematic distortions are corrected by establishing mathematical relations between the coordinates of pixels in an image and the corresponding coordinates of those points on the ground. These relations can be used to correct the image geometry irrespective of prior information about the source and type of distortion. The second approach is the most commonly used and it is independent of the platform used for image acquisition. This approach involves two steps. First step is registration process. It involves identifying raw image coordinates (i.e. row, column) of several clearly discernible points, called ground control points (GCPs), in the distorted image, and map them either to their true positions in ground coordinates (e.g. latitude, longitude) measured from a map (image-to-map registration), or to georeferenced image (corrected before), coordinates of corresponding points (image-toimage registration), through to mathematical transformation, that will convert the raw image coordinates into the desired coordinates [3]. The second step is the resampling process to recalculate the gray level values for pixels in the transformed output image based on pixel values in the input (uncorrected) image. There are three common methods for resampling: nearest neighbor, bilinear interpolation, and cubic convolution [5]. Many studies have been carried out on the geometric correction of satellite images, either using the first approach, [6], [7], [8], [9], [10], or using the second approach [11], [12], [13], [14]. 2/14

3 2- Satellite images data under study In this paper the effect of the selected location of GCPs and the way of their distribution over the distorted image area on geometric correction accuracy was applied on a sample of two raw (source) satellite images, named image1and image2. These two images are acquired in 1999 by remote sensing satellite LANDSAT 5. They are characterized by the following: a. Image1 and image2 represent the areas of north delta and south delta in Egypt respectively. b. Image1 and image2 are acquired in red ( μm), green ( μm), and blue ( μm) spectral bands. c. The size of image1 is 490 x 697 pixels and the size of image2 is 745 x 939 pixels. Two geometrically corrected mages, of the same area, size, and imaging band, are used as reference images. The reference images are acquired in 1999 by LANDSAT 5. They were geometrically corrected using maps (image-to- map) registration. 3- Geometric correction process using mapping polynomials Mapping polynomials correction approach depends on establishing mathematical relationships (mapping function) between the addresses of pixels in the distorted image and the corresponding coordinates of those points on the ground via a map or pre-corrected image, represented generally as follows: u = f (x, y) where: v = g (x, y) x, y.the location of points in the reference image (or map in image-to-map registration) u, v the location of pixels in the source (distorted) input image f, g..the pair of mapping functions. (1) In fact explicit forms for the mapping functions (1), are not known therefore they are generally chosen as simple polynomials of different degrees. The polynomial equations for a t-degree transformation take this form [4]: where: u v = = u and v are source coordinates (distorted image). x and y are reference coordinates. t is the order (degree) of the polynomial a k and b k are the polynomial coefficients 3/14 t t i = 0 i = 0 i i j = 0 j = 0 a b k k x x i i j j y y j j (2)

4 The minimum number of selected GCPs depends on the polynomial order. The relations between the order of the polynomials; the minimum required number of GCPs; and number of coefficients of the used polynomials, are given by [5]. ( t + 1 )( t + 2 ) (3) N = 2 where: N t M M minimum number of GCPs order of the polynomial number of coefficients ( t + 1 )( + 2 ) = t In this work a polynomial of third order is applied, this means that the minimum required number of GCPs is 10 points. (4) 4- Experimental work The experimental work is conducted using Earth Resources Data Analysis System (ERDAS) Imagine 8.5 Software. A polynomial of 3 rd order (10 GCPs) is used for image-to-image registration. The nearest neighbor resampling method is used to calculate the pixels gray level values of the rectified output image. The accuracy of the correction process is evaluated by calculating the RMS error at every GCP. The RMS error is the difference between the desired output coordinate for a GCP and the actual output coordinate for the same point, when the point is transformed with the geometric transformation. RMS error in X, Y directions and total (T) RMS error at the GCPs are calculated according to the following equations [5]: RMS (in X) = RMS (in Y) = RMS (T) = 1 n i = 1 n i = 1 n i n = 1 ( ΔX ) i ( ΔY ) i ( ΔX i + ΔY i ) (5) where: ΔX i, ΔY i T n i = residuals of point ( i ) in X and Y directions. = total RMS error = number of GCPs = GCP number 4/14

5 5- Results and analysis By a good and a bad location of a selected GCP, we mean it is identified on road intersections, airport runway intersections, bends in rivers features and the like, or it is identified on the others (no clear describable features) respectively. For a good distribution of GCPs, we take into consideration the polynomial order, starting from 1 st order. It means that the first 3 GCPs (min number in case of 1 st order) should be distributed far away from each other. The bad distribution is the opposite, i.e first 3 GCPs are distributed near to each other. Figure 1 (a,b) shows the raw (source) image1, and its reference image. Figure 1 (c,d) shows a good location and bad distribution of the selected 10 GCPs in both source and reference image. Figure 1 (e,f) shows the corrected (output) image and its reference respectively. Table (1) gives the calculated average RMS error value (0.3). Figure 2 (a,b) shows the raw (source) image1, and its reference image. Figure 2 (c,d) shows a good location and good distribution of the selected 10 GCPs in both source and reference image. Figure 2 (e,f) shows the corrected (output) image and its reference respectively. Table (2) gives the calculated average RMS error value (0.2) Figure 3 (a,b) shows the raw (source) image1, and its reference image. Figure 3 (c,d) shows a bad location and good distribution of the selected 10 GCPs in both source and reference image. Figure 3 (e,f) shows the corrected (output) image and its reference respectively. Table (3) gives the calculated average RMS error value (0.78) The raw (source) image2 and its reference image are shown in Fig.4 (a,b). Figure 4 (c,d) shows a good location and bad distribution of the selected 10 GCPs in both source and reference image. The corrected (output) image and its reference are given in Fig.4 (e,f) respectively. Table (4) gives the calculated average RMS error value (0.11). The raw (source) image2 and its reference image are shown in Fig.5 (a,b). Figure 5 (c,d) shows a good location and good distribution of the selected 10 GCPs in both source and reference image. The corrected (output) image and its reference are given in Fig.5 (e,f) respectively. Table (5) gives the calculated average RMS error value (0.05). The raw (source) image2 and its reference image are shown in Fig.6 (a,b). Figure 6 (c,d) shows a bad location and good distribution of the selected 10 GCPs in both source and reference image. The corrected (output) image and its reference are given in Fig.6 (e,f) respectively. Table (6) gives the calculated average RMS error value (0.4). The raw (source) image2 and its reference image are shown in Fig.7 (a,b). Figure 7 (c,d) shows a bad location and good distribution of the selected 10 GCPs in both source and reference image. The corrected (output) image and its reference are given in Fig.7 (e,f) respectively. Table (7) gives the calculated average RMS error value (0.71). 5/14

6 Table (1) Average of RMS error of image 1 No. of GCPs First Second Third Table (2) Average of RMS error of image 2 No. of GCPs First Second Third Table (3) Average of RMS error of image 2 No. of GCPs First Second Third Table (4) Average of RMS error of image 2 No. of GCPs First Second Third Table (5) Average of RMS error of image 1 No. of GCPs First Second Third Table (6) Average of RMS error of image 1 No. of GCPs First Second Third Table (7) Average of RMS error of image 2 No. of GCPs First Second Third /14

7 6- Conclusion The following points conclude the whole paper: a. The bad location and bad distribution of the selected GCPs lead to increase in the average RMS error value of correction of an image b. The effect of bad location of selected GCPs is more sever than that of bad distribution of selected GCPs on the correction accuracy. c. To obtain high accuracy of geometric correction of remote sensing satellite images, the location and distribution of selected GCPs should be taken into consideration as mentioned before. 7- References [1] Fawzy Eltohamy, Gouda Ismail, Essam Hamza, H. Hussien " Geometric Correction of Remote Sensing Satellite Digital Images Using Mapping Polynomial of Different orders" [2] Environmental Remote Sensing Forestry 753 Lab Three: Geometric Correction, (2005). [3] John Richards & Xiuping Jia Remote Sensing Digital Image Analysis An Introduction Springer, 3 rd edition,(1999). [4] Thomas M. Lillesand & Ralph W. Kiefer Remote Sensing and Image Interpretion 4 th edition, John Wiley & Sons, (2000). [5] Erdas Imagine 8.5 Field Guide, (2002). [6] Dowmann I., and Dolloff, J., An evaluation of rational function for Photogrammetric restitution, International Archives of Photogrammetry and Remote Sensing, Amsterdam, The Netherlands, July 16-23, (Amsterdam, The Netherlands: GITC) Vol. 33 (B3), pp (2000). [7] El-Manadili, Y., and Novak, K.,. Precision Rectification of spot Imagery using the direct linear [8] Kratky, V Rigorous Photogrammetric Processing of Spot Images at CCM Canada, ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 44, pp (1989). [9] Konecny, G., Lohmann, P., Engel, H., and Kruck, E, Evaluation of SPOT Imagery on Analytical Instruments, Photogrammetric Engineering and Remote Sensing, 53, , (1987). [10] Xutong Niu, Jue Wang, Kaichang Di, Jin-Duk Lee, Ron Li,. Geometric Modelling and Photogrammetric Processing of High Resolution Satellite Imagery Mapping and GIS Laboratory, CEEGS, The Ohio State University. (2004). [11] Hafez Abbas Afify,. Planimetric Accuracy of Rectified Spot Imagery Lecturer, Dpt. Of Transportation Engineering, Faculty of Engineering, Tanta University, Tanta, Egypt. (2002). [12] A. M. Wu and Y. Y. Lee. Geometric Correction of High Resolution Images Using Ground Control Points Singapore Institute of Surveyors and Valuers (SISV); Asian Association on Remote Sensing (AARS). (2001). [13] Yoshinari OGURO, Shoji TAKEUCHI and Yuzo SUGA, An Improvement in Educational Tools For Geometric Correction of Satellite Images by Using DEM Shade Simulation.(2002). [14] Thierry Toutin,. Review Paper: Geometric Processing of Remote Sensing Images: Models, Algorithms and Methods, Canada Centre for Remote Sensing. (2003). 7/14

8 (a) (b) (c) (d) (e) (f) Fig. 1. a & b Raw image1 and its corresponding reference image, c & d 10 GCPs good location & bad distribution on image1 & its reference, e & f Corrected image1, and its reference with link between them. 8/14

9 (a) (b) (c) (d) (e) (f) Fig. 2. a & b Raw image1 and its corresponding reference image, c & d 10 GCPs good location & good distribution on image1 & its reference, e & f Corrected image1, and its reference with link between them. 9/14

10 (a) (b) (c) (d) (e) (f) Fig. 3. a & b Raw image1 and its corresponding reference image, c & d 10 GCPs bad location & good distribution on image1 & its reference, e & f Corrected image1, and its reference with link between them. 10/14

11 (a) (b) (c) (d) (e) (f) Fig. 4 a & b Raw image2 and its corresponding reference image, c & d 10 GCPs good location & bad distribution on image2 & its reference, e & f Corrected image2, and its reference with link between them. 11/14

12 (a) (b) (c) (d) (e) (f) Fig. 5 a & b Raw image2 and its corresponding reference image, c & d 10 GCPs good location & good distribution on image2 & its reference, e & f Corrected image2, and its reference with link between them. 12/14

13 (a) (b) (c) (d) (e) (f) Fig. 6 a & b Raw image2 and its corresponding reference image, c & d 10 GCPs bad location & good distribution on image2 & its reference, e & f Corrected image2, and its reference with link between them. 13/14

14 (a) (b) (c) (d) (e) (f) Fig. 7 a & b Raw image2 and its corresponding reference image, c & d 10 GCPs bad location & bad distribution on image2 & its reference, e & f Corrected image2, and its reference with link between them. 14/14