With the increase of the spatial resolution of aerial images, the shadow problem is more prominent. The shadows affect the applications such as object recognition, image dense matching, and object classification. Existing shadow detection methods can acquire results with high accuracy, but usually need much manual intervention and the traditional stretch models generally lead to color distortion and undercompensation. Therefore, we propose an object-oriented automatic shadow detection method without manual intervention and a shadow compensation method by regional matching. In the proposed method, pixel-based soft shadow detection, which uses Gaussian mixture model to simulate the gray distribution and refines soft shadow map with guiled filtering, is combined with image segmentation result to obtain accurate shadow regions with complete shape and no hole. Then shadow regions are compensated, with less loss of details and brightness imbalance, referring to their optimal homogeneous nonshadow region obtained by regional matching based on Bag-of-Words. The total variation model is used to decrease the noise amplified by the pixel-based stretch and boundary effect in the result. Experiments are performed on three publicly available high-resolution aerial images to demonstrate the superiority of our proposed methods. It shows that the proposed method can accurately detect shadows from urban high-resolution aerial images with an overall rate of over 88%. The compensation results display excellent visual effects compared with the state-of-the-art methods, consistent with the true color of ground objects.
The most commonly used features in unsupervised change detection are spectral characteristics. Traditional methods describe the degree of the change between two pixels by quantifying the difference in spectral values or spectral shapes (spectral curve shapes). However, traditional methods based on variation in spectral shapes tend to miss the change between two pixels if their spectral curves are close to flat; and traditional methods based on variation in spectral values tend to miss the change between two pixels if their values are low (dark objects). To inhibit the weaknesses of traditional methods, a novel approach to unsupervised change detection based on hybrid spectral difference (HSD) is proposed which combines the difference between spectral values and spectral shapes. First, a new method referred to as change detection based on spectral shapes (CDSS) is proposed that fuses the difference images produced by the spectral correlation mapper (SCM) and spectral gradient difference (SGD) in order to describe the variation in spectral shapes. Second, a method called change detection based on spectral values (CDSV), computing the Euclidean distance between two spectral vectors, is used to obtain a difference image based on the variation in spectral values. Then, the credibility of CDSS and CDSV for every pixel is calculated to describe how appropriate these two methods are for detecting the change. Finally, the difference images produced by CDSS and CDSV are fused with the corresponding credibility to generate the hybrid spectral difference image. Two experiments were carried out on worldview-2/3 and Landsat-7 Enhanced Thematic Mapper Plus (ETM+) datasets, and both qualitative and quantitative results indicated that HSD had superior capabilities of change detection compared with standard change vector analysis (CVA), SCM, SGD and multivariate alteration detection (MAD). The accuracy of CDSS is higher than CDSV in case-1 but lower in case-2 and, compared to the higher one, the overall accuracy and the kappa coefficient of HSD improved by 3.45% and 6.92%, respectively, in the first experiment, and by 1.66% and 3.31%, respectively, in the second experiment. The omission rate dropped by approx. 4.4% in both tests.