Shadow detection improvement using spectral indices and morphological operators in high resolution images from urban areas

Abstract

ABSTRACT:While high-resolution remote sensing images have increased application possibilities for urban studies, the large number of shadow areas has created challenges to processing and extracting information from these images. Furthermore, shadows can reduce or omit information from the surface as well as degrading the visual quality of images. The pixels of shadows tend to have lower radiance response within the spectrum and are often confused with low reflectance targets. In this work, a shadow detection method was proposed using a morphological operator for dark pattern identification combined with spectral indices. The aims are to avoid misclassification in shadow identification through properties provided by them on color models and, therefore, to improve shadow detection accuracy. Experimental results were tested applying the panchromatic and multispectral band of WorldView-2 image from São Paulo city in Brazil, which is a complex urban environment composed by high objects like tall buildings causing large shadow areas. Black top-hat with area injunction was applied in PAN image and shadow identification performance has improved with index as Normalized Difference Vegetation Index (NDVI) and Normalized Saturation-Value Difference Index (NSDVI) ratio from HSV color space obtained from pansharpened multispectral WV-2 image. An increase in distinction between shadows and others objects was observed, which was tested for the completeness, correctness and quality measures computed, using a created manual shadow mask as reference. Therefore, this method can contribute to overcoming difficulties faced by other techniques that need shadow detection as a first necessary preprocessing step, like object recognition, image matching, 3D reconstruction, etc.* Corresponding author. This is useful to know for communication with the appropriate person in cases with more than one author.

Snake-Based Model for Automatic Roof Boundary Extraction in the Object Space Integrating a High-Resolution Aerial Images Stereo Pair and 3D Roof Models

“…In order to generate this product, shadows should be detected in aerial images. We used the shadow detection algorithm proposed by Azevedo et al [21], but with some adaptations as presented below.…”

“…Before applying the method proposed by Azevedo et al [21], we perform a smoothing process to the original image. This is usually necessary because the employed images are of high resolution; as a result, the shadow detection method can identify many unnecessary shadow details caused by small objects, such as the roof tiles.…”

“…This transformation detects the dark patterns in the image through the arithmetic difference between the original image and its closing. As the shadows in aerial images do not obey a specific pattern, the original work by Azevedo et al [21] uses the area closing operation so that there is no need to establish a particular shape for the structuring element. However, the type of shadow of interest in this work is those cast by the buildings themselves, having rectangular and elongated shapes.…”

Object-Based Approach Using Very High Spatial Resolution 16-Band WorldView-3 and LiDAR Data for Tree Species Classification in a Broadleaf Forest in Quebec, Canada

“…Finally, the images were mosaicked with a bundle adjustment and then fitted with the CHM using 100 tie points to reduce the offset in the canopy [83]. A second-order polynomial regression was used to create the final mosaic with a root mean square error of 0.97 m. Inspired by Zhou and Qiu [84] and Hartling et al [31], deep shadow was extracted from the mosaic using a maximum likelihood classification with a shadow index [85]. The Bhattacharyya index showed a separability of 1.94 to detect deep shadow against other elements.…”

Shadow detection using object area-based and morphological filtering for very high-resolution satellite imagery of urban areas

“…In addition, as a connected filter, the area closing operator only acts on the flat zones of the image (a set of connected iso-intensity pixels), avoiding shape distortion and the loss of spatial characteristics of the objects of interest. 35 Azevedo et al 36 set the area parameter empirically, making their approach semiautomatic instead of nonsupervised. This parameter depends on the resolution of the target images so that the connectivity of the image objects must be large enough when coping with VHR images.…”