International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 5 - November 2015 Impact of the Integrability constraint on Radar Shape-fromShading Mobarak, Babikir A. Civil Engineering Dept., Faculty of Engineering, Al-Baha University Al baha, Saudi Arabia Abstract — In Synthetic Aperture Radar (SAR) remote sensing, DEMs are used to resample the SAR images to well known coordinate systems. Also, they are used to correct for distortions originated from imaging sensor. Radar Shape from Shading (RSFS) is one of different techniques that used to extract heights from SAR data sets. The most important key to this technique is enforcing of the integrability constraint in frequency-domain. Fourier transformation is used to transforms the data from time-domain to frequencydomain. The performance of this constraint was evaluated on RADARSAT-1 image using both graphical and statistical analysis. To show the effect of integrability constraint, the RSFS technique was evaluated without and with enforcing the integrability. The RMSE and R2 were found to be 31.98m and 0.705 and 17.47m and 0.972, respectively, showing the effect of the constraint. Keywords — SFS, DEM, Integrability, SFS Constraints. SAR Imageries, I. INTRODUCTION A. SHAPE FROM SHADING Point density or accuracy of DEMs generated from direct techniques such as ordinary surveying, photogrammetry, and remote sensing for some parts of the Earth, such as a tropical area, is still insufficient for some applications. This is due to some difficulties resulted from the nature of these areas or the limitation of the technique or system used. Shape-from-shading (SFS) technique deals with the recovery of shape from a gradual variation of shading in the image [2]. The classic assumption underlining SFS is that the surface under investigation has a Lambertian reflectivity. The first systematic study of SFS was reported by [5] and his colleagues [8]. SFS deals with the process of finding the object’s 3D shape from a single image of that object. The use of a single image cannot always ensure the uniqueness of the shape of an object. Therefore, there will be relatively little effect devoted to exploiting the exact 3D shape reconstruction from the shading information of one image. This problem is resolved by introducing ancillary information to the SFS process. The basic assumption underlying SFS is a uniform surface reflectivity (Lambert). Several studies investigating Lambertian reflectance model have been carried out on SFS ([7], [13], and [10]). ISSN: 2231-5381 From a computational viewpoint, SFS involves solving the image irradiance equation to recover a set of surface normals or surface slopes [14]. Reference [5] was the first researcher, who had formulated SFS problem and found the solution as a nonlinear firstorder partial differential equation (PDF). This equation is known as the image irradiance equation and is the basic equation for any SFS technique. It relates the image irradiance to the scene radiance as shown in Equation 1 below: E ( x,y) ˆ ( x,y)) R( n (1) Where E(x,y) is the image irradiance at a point (x,y), R is the reflectivity, and n̂ represents the three components of unit surface normal. The recovered surface can be expressed in four types [3]; surface height (elevation) z(x,y), surface normal (nx, ny, nz), surface slope (p,q), and surface slant Φ and tilt ϴ . The depth can be considered either as the relative distance from the camera or antenna to the surface points, or the relative surface height above the xy plane. This implies that equation 1 can also be written as follows: E ( x,y) R( p, q) (2) Where (p,q) = (dz/dx , dz/dy) B. RADAR SHAPE FROM SHADING SAR sensor can obtain remote sensing data in all weather conditions due to its capability of penetrating clouds and working day or night. The singular nature of active radar instruments allow microwaves to interact with surface features, means that information obtained can be dependent on moisture content, salinity and physical characteristics such as shape, size, and orientation [6]. There are many difficulties in acquiring images by conventional optical sensors, especially in cloud-covered regions and/or season. Currently, there are new techniques developed for height extraction from SAR data. These include radar stereogrammetry, microwave altimetry, SAR interferometry, and Light Detection and RangingLIDAR [11]. Point density or accuracy of DEMs generated from direct techniques such as ordinary surveying, photogrammetry, and remote sensing for some parts of the Earth, such as a tropical area, is still insufficient for some applications. This is due to some http://www.ijettjournal.org Page 251 International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 5 - November 2015 difficulties resulted from the nature of these areas or the limitation of the technique or system used. The major objection to RSFS is the ambiguity from uncertain backscatter properties. Involving some constraints like brightness, smoothness, and integrability can remove or reduce this ambiguity. C. INTEGRABILITY This constraint is important if the surface is to be reconstructed from the recovered field of “surface normal” [12]. This is due to the fact that the estimated height will depend on the integration path if the constraint is not satisfied. The integrability constraint ensures valid surfaces, that is p y q x . It ensures that heights can be integrated along any path, because these values are independent of path of integration. Enforcing this constraint practically smoothens the estimated surface. This condition can be described by ( p y qx ) 2 dxdy (3) Reference [4] proposed a method for enforcing integrability in [1] algorithm by projecting non integrable surface slope estimates onto nearest integrable surface slopes. One way to obtain such projection is by Fourier transformation. Their results showed that the accuracy and efficiency had improved over the Brooks and Horn’s algorithm. II. MATERIALS AND METHODOLOGY Figure 1: Subset from RADARSAT-1 S7 Mode Image c. Methodology Digital image pre-processing was carried out to prepare the RADARSAT-1 SAR image first. Then the geometric correction was done to RADARSAT-1 image using some GCPs. After that the Radar brightness (β0) and backscatter coefficients (σ◦) were calculated using equations 4 and 5, respectively. βor,a =10*log10((DNr,a2 + A3) / A2r) σor,a o =β r,a +10*log10(sin θ r) (4) (5) Where: DNr,a is a digital number at range (r) and azimuth (a), The study area is the KASSALA state in the east of A3 is a small constant (often 0), SUDAN. It lies between longitudes 35.59693ºE and A2r is a range dependent look-up table that contains a 36.44708ºE and latitudes 15.11390ºN and 16.17869ºN. terrain type model. The most interesting features included within the area θr is the incident angle at range (r). of study are ALGASH River and the TAKAH Then, speckle filtering was applied to the Mountain. backscatter coefficients to remove (or reduce) the speckle “inherent with radar data” from the image. b. Data Enforcing of the Integrability The data sets for this study consist of one Integrability constraint was enforced after RADARSAT-1 image, covering the study area, and transformation of the data from time-domain to ground control points (GCPs). A subset from image of frequency-domain using Fourier transformation. The RADARSAT-1 was extracted to examine the integrability was enforced within the radar SFS model performance of the integrability. The area of the proposed by [9]. Integrability requires that the subset was approximately 10 km2. Figure 1 below recovered surface height at any particular point is represents a subset obtained from RADARSAT-1 S7 independent of the direction of integration and is mode image. defined as expressed in Equation 3. a. Area of Study p y or ISSN: 2231-5381 http://www.ijettjournal.org q x py (6) qx Page 252 International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 5 - November 2015 Integrability requires that the recovered surface height at any particular point is independent of the direction of integration and is defined as expressed in Equation 3.4. ~ ( )Z ( ) ~ ( )Z ( ) y ~ p( ) ~( ) q a a (7) x where, can be computed using the transformation equation below ~ Z( ) a cnj x cnj ˆ( ) ( )p a a 2 ax ( ) y ( )qˆ ( ) 2 ( ) y where, ) is the Fourier transform of the integrable height Z(x,y), p̂ ( ) and q̂( ) are the Fourier transform of the surface slope estimates in x and y-direction, a x ( ) and a y ( ) are the Fourier coefficients of a discreet differentiation operator in x and y, cnj a x ( ) and cnj a x ( ) are the conjugates of a x ( ) and a y ( ) , and ω = (ωx, ωy) represents the two-dimension frequency coordinates. Equation 8 minimizes the following distance simultaneously as shown in Equation 9, while satisfying the conditions expressed by Equation 6. d ( p̂, q̂) , ( ~ p , q~ ) 2 ~ p p̂ 2 q~ q̂ dxdy (9) Fourier transform of ( pˆ ,qˆ ) was performed to get (p̂( ), q̂( )) and then substitution of them into Equation 8 was done to obtain the Fourier transforms ~ of integrable surface height Z ( ) . Enforcing the integrability constraint was carried out by calculating the Fourier transforms of the nearest integrable surface slopes (~ p( ), ~ q( )) using Equation 10 below ~ ax ( ) Z ( ) ~ ay ( )Z ( ) ~ p( ) ~( ) q (10) Where, ax j sin( ( In the last step inverse Fourier transforms of slope estimates (~ p( ), ~ q( )) was performed to obtain the new integrable surface slopes ( p ˆ ,qˆ ) . III. RESULTS AND DISCUSSION ~ Z( ) ~ Z( are the input image row and column numbers, respectively. x , y x ) ) , ay j sin( y ), j 1 , ( 2 m / M ,2 n / N ) , m = - M/2,…,0,…,M/2 , n = -N/2,…,0,…,N/2, and M and N ISSN: 2231-5381 The objective of this section of study was to assess the importance of the integrability constraint on surface heights recovery using RSFS technique. The integrability constraint was enforced as a projection (8) constraint. An integrable set of surface slope estimates was constructed from non integrable ones. In the case of the developed SFS algorithm operated without the integrability constraint, any number of surface height reconstructions could be produced depending upon the path of integration. Consequently, the surface reconstructed could be noisy if the surface slope estimates were noisy. The performance of the integrability constraint was tested on RADARSAT-1 image of Standard mode (S7). The evaluation of the accuracy was based on the comparison of absolute surface heights derived by the enforcing of the integrability and the real height obtained from GCPs. Evaluation was carried out qualitatively and quantitatively. Qualitative evaluation was conducted by investigating the shape of surface topography reconstructions and their actual values in the study area. Quantitative evaluation was done statistically through calculations of RMSE and R2. Some 123 GCPs were used for this purpose. It was considered that quantitative measures would supplement and extend the qualitative analysis. Figures 2 and 3 represent the resultant image of plotting absolute heights, reconstructed from RSFS algorithm without and with enforcing the integrability constraint, respectively. The most interesting observation is that there are significant differences between the unconstrained (Figure 2) and constrained model results (Figure 3). Few obvious distortions are observed in Figure 2. The overall surface height values range between 425m and 800m. In addition, the location of the peak of Taka mountain is shifted to the lower-left corner. The value of this point, as estimated from the figure, is approximately 800m. Referring to Figure 4, it is observed that the range between minimum and maximum surface heights of non-integrable surface height is smaller than that of the integrable one (400m-1009m). Also, it is clear that the reconstructed surface height produced by enforcing the integrability constraint is smoother than that obtained without enforcing the integrability constraint. Table Error! No text of specified style in document.1 shows the RMSE and R2 results of surface heights recovered for both the constrained and unconstrained RSFS algorithms. It is clear that the integrability http://www.ijettjournal.org Page 253 International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 5 - November 2015 constraint reduces the RMSE and R2 respectively from 31.98m to 17.47m and from 0.705 to 0.972, giving the differences of 14.51 and 0.267. Table Error! No text of specified style in document.1: RMSE and R2 of Reconstructed Surface Height from Constrained and Unconstrained SFS algorithm Model Unconstrained Constrained RMSE 31.98 17.47 0.705 0.972 R 2 The finding of integrability constraint is consistent with that of [8], who found that most of surface smoothing came from integrability constraint. IV. CONCLUSION Figure 2: Plot of Absolute Height from SFS without the Integrability Constraint It is interesting to note that enforcing the integrability constraint has a significant effect on the final absolute height accuracy after comparison with the results obtained without enforcing this constraint. The validation of the later results gave RMSE and R2 of 31.98m and 0.705, respectively. This finding indicated that most of surface smoothing was attributed to the integrability constraint. REFERENCES [1] [2] [3] [4] Figure 3: Plot of Absolute Height from SFS with the Integrability Constraint [5] [6] [7] [8] [9] [10] [11] Figure Error! No text of specified style in document.: Height Differences between Constrained and Unconstrained Model [12] [13] ISSN: 2231-5381 Brooks, M. J., and Horn, B. K. P. Shape and source from shading: Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts , United States, 1985. Du, Q. Y., Chen, S. B., and Lin, T. An application of shape from shading. 2004 8th International Conference on Control, Automation, Robotics and Vision (ICARCV), 2004 pp. 184189. Durou, J. D., Falcone, M., and Sagona, M. Numerical methods for shape-from-shading: A new survey with benchmarks. ELSEVIER, Computer Vision and Image Understanding, 2008, vol.109(1), pp: 22-43. Frankot, R. T., and Chellappa, R. A Method for Enforcing Integrability in Shape from Shading Algorithms. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1988, Vol. 10(4), pp. 439 - 451. Horn, B. K. P. "Obtaining shape from shading information," in The Psychology of Machine Vision 1975. Iain, H. Woodhouse. Introduction to Microwave Remote Sensing. University of Edinburgh, Scotland, UK: Taylor & Francis 2005. Kimmel, R., and Bruckstein, A. M. Shape from Shading via Level Sets (CIS Report No. 9209): Israel Institute of Technology 1992. Liu, Hongxing. Derivation of surface topography and terrain parameters from single satellite image using shape-fromshading technique. Computer & Geosciences, 2003, Vol. 29(10), pp.1229-1239. Mobarak, B. A., Mansor, S. B., Shariff, A.R. M. Bejo, K.N. Bt. 1 and Mohamed, N.K. Generation of DEMs from Single SAR Image Using Radar Shape from Shading Technique. International Geoinformatics, Research and Development Journal (IGRDJ), Vol. 1 (1); pp. 59-69. 2010. Prados, E., and Faugeras, O. Shape from shading: A wellposed problem. IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2, 2005, p.p. 870877. Rajabi, M. A. Spatial Enhancement Of Digital Terrain Models Using Shape From Shading With Single Satellite Image. The University Of Calgary, CALGARY ALBERTA, CANADA. 2003. Robles-Kelly, A., and Hancock, E.R. A graph-spectral approach to shape-from-shading IEEE Transactions on Image Processing 2004, Vol. 13(7), pp. 912-926. Wilson, R. C., and Hancoc, E. R. A radar reflectance model for terrain analysis using shape from shading. IEEE http://www.ijettjournal.org Page 254 International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 5 - November 2015 [14] [15] [16] International Conference on Image Analysis and Processing, 1999 pp.868-873. Worthington, P. L. New constraints on data-closeness and needle map consistency for shape-from-shading. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1999 Vol. 21(12), pp. 1250 -126 Shibam Das , Ambika Aery. "A Review : Shadow Detection And Shadow Removal from Images". International Journal of Engineering Trends and Technology (IJETT). V4(5):17641767 May 2013. ISSN:2231-5381. Vivek Arya, Dr. Priti Singh , Karamjit Sekhon. "RGB Image Compression Using Two Dimensional Discrete Cosine Transform". International Journal of Engineering Trends and Technology (IJETT). V4(4):828-832 Apr 2013. Dr. Babikir A. Mobarak received his B. Tech. and M.Sc. degrees in Surveying Engineering and Geodetic Survey from Sudan University of Science and Technology (SUST), Khartoum, SUDAN in 1996 and 2003, respectively, and the Ph.D. degree in GIS and Geomatic Engineering from Universiti Putra Malaysia (UPM), Serdang, Malaysia, in 2011. From 1997 to 2003, he was TA, in Surveying Eng. Dept. College of Eng. (SUST). From 2003 to 2005, he was also in Surveying Eng. Dept. College of Eng. (SUST) as a Lecturer. From 2011 to 2014 he was Assistant Professor, School of Surveying, College of Eng. (SUST). Now he is an Assistant Professor in Civil Engineering Department, Faculty of Engineering, Al Baha University, Al Baha, Saudi Arabia. ISSN: 2231-5381 http://www.ijettjournal.org Page 255