3440 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 9, SEPTEMBER 2012 A Complete Processing Chain for Shadow Detection and Reconstruction in VHR Images Luca Lorenzi, Student Member, IEEE, Farid Melgani, Senior Member, IEEE, and Grégoire Mercier, Senior Member, IEEE Abstract—The presence of shadows in very high resolution (VHR) images can represent a serious obstacle for their full ex- ploitation. This paper proposes to face this problem as a whole through the proposal of a complete processing chain, which relies on various advanced image processing and pattern recognition tools. The first key point of the chain is that shadow areas are not only detected but also classified to allow their customized compensation. The detection and classification tasks are imple- mented by means of the state-of-the-art support vector machine approach. A quality check mechanism is integrated in order to reduce subsequent misreconstruction problems. The reconstruc- tion is based on a linear regression method to compensate shadow regions by adjusting the intensities of the shaded pixels according to the statistical characteristics of the corresponding nonshadow regions. Moreover, borders are explicitly handled by making use of adaptive morphological filters and linear interpolation for the prevention of possible border artifacts in the reconstructed image. Experimental results obtained on three VHR images representing different shadow conditions are reported, discussed, and com- pared with two other reconstruction techniques. Index Terms—Image enhancement, image restoration, missing data, shadow detection, shadow reconstruction, support vector machines (SVMs), very high resolution (VHR) images. I. I NTRODUCTION R ECENTLY, very high resolution (VHR) satellite images opened a new era in the remote sensing field. Because of the increase of spatial resolution, new analysis, classification, and change detection techniques are required. Indeed, VHR images exhibit resolutions which allow distinguishing very well detailed features from small objects, like little building structures, trees, vehicles, and roofs. Unfortunately, high spa- tial resolution entails also some drawbacks like the unsought presence of shadows, particularly in urban areas where there Manuscript received February 22, 2011; revised October 21, 2011; accepted November 25, 2011. Date of publication March 5, 2012; date of current version August 22, 2012. L. Lorenzi is with the Department of Information Engineering and Computer Science, University of Trento, 38123 Trento, Italy and also with the Image and Information Processing Department, Institut Telecom, Telecom Bretagne, 29238 Brest, France (e-mail: lorenzi@disi.unitn.it; luca.lorenzi@telecom- bretagne.eu). F. Melgani is with the Department of Information Engineering and Com- puter Science, University of Trento, 38123 Trento, Italy (e-mail: melgani@ disi.unitn.it). G. Mercier is with the Image and Information Processing Depart- ment, Institut Telecom, Telecom Bretagne, 29238 Brest, France (e-mail: gregoire.mercier@telecom-bretagne.eu). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TGRS.2012.2183876 are larger changes in surface elevation (due to the presence of buildings, bridges, towers, etc.) and consequently longer shad- ows. Although it is feasible to exploit shadow characteristics to recognize building position and to estimate their height and other useful parameters [1], [2], usually, shadows are viewed as undesired information that strongly affects images. Shadows may cause a high risk to present false color tones, to distort the shape of objects, to merge, or to lose objects. They represent an important problem for both users and sellers of remote sensing images [2]. As a consequence, shadows can impact neg- atively in the exploitation of VHR images, influencing detailed mapping, leading to erroneous classification or interpretation (e.g., biophysical parameters such as vegetation, water, or soil indexes), due to the partial or total loss of information in the image [3]. To attenuate these drawbacks and, thus, to increase image exploitability, two steps are necessary: 1) shadow detec- tion and 2) shadow compensation (reconstruction). An example of the importance of getting shadow-free images is the massive tsunami in 2004 where it was crucial to obtain such images in a very short time in order to take rapid and crucial decisions in rescue missions [4]. The literature reports mainly two approaches to detect shadows, namely, model-based and shadow-property-based ap- proaches. The former needs prior information about the sce- nario and the sensor. However, since, usually, such knowledge is not available, most of the detection algorithms are based on shadow properties, such as the fact that shadow areas have lower brightness, higher saturation, and greater hue values [5]. For instance, methods in [6] and [7] attempt to detect shadows using a space color transformation and an automatic threshold estimator (e.g., Otsu’s algorithm [8]). In a comparative work [9], several invariant color spaces were analyzed to detect shad- ows, namely, hue, intensity, and saturation (HIS), hue saturation value (HSV), HCV, YIQ, and YCbCr transforms. Inspired by this comparative analysis, a better approach was developed, which is based on a novel successive thresholding scheme [10]. Other algorithms rely on the idea of adding features capable to better discriminate shadow areas (e.g., normalized difference vegetation index [11], normalized saturation-value difference index (NSVDI) [12], and maximally stable extremal regions [13]). Another technique applies the principal compo- nent analysis to isolate the luminance component in an RGB image, where the detection of shadows appears more accurate [14]. Finally, physical properties (e.g., temperature) of a black- body radiator have been exploited in a recent method to detect shadows [15]. 0196-2892/$31.00 © 2012 IEEE