Jorge Lira

Segmentation and morphology of open water bodies from multispectral images

A number of problems in remote sensing require the segmentation of specific spectral classes such as water bodies, clouds or forested areas. Further analysis of these classes may include the calculation of optical reflectance values such as chlorophyll concentration, absolute reflectivity or vegetation indices. To derive reliable measurements of these variables, a precise segmentation − from the rest of the image − of the spectral classes is needed. In this work, we propose a new methodology to segment open water bodies based on a variant of principal component analysis (PCA). In this variant, information about the spectral class model of the water bodies to be separated in the feature space is required. This information is input by means of a training field encircling a set of pixels representative of this spectral model. A training field for land cover is also defined. This PCA variant produces two sets of multispectral bands, one for water bodies and one for land cover types. The first two bands of each set are input into a fuzzy clustering procedure. By using a merging process, the clusters are merged into two classes: water bodies and the rest of the image. From this, a logic bitmap image is obtained. The pixels of the bitmap consist of ON for water bodies and OFF for the rest of the image. The bitmap is then used to obtain morphological parameters of the water bodies.

A supervised contextual classifier based on a region-growth algorithm

A supervised classification scheme to segment optical multi-spectral images has been developed. In this classifier, an automated region-growth algorithm delineates the training sets. This algorithm handles three parameters: an initial pixel seed, a window size and a threshold for each class. A suitable pixel seed is manually implanted through visual inspection of the image classes. The best value for the window and the threshold are obtained from a spectral distance and heuristic criteria. This distance is calculated from a mathematical model of spectral separability. A pixel is incorporated into a region if a spectral homogeneity criterion is satisfied in the pixel-centered window for a given threshold. The homogeneity criterion is obtained from the model of spectral distance. The set of pixels forming a region represents a statistically valid sample of a defined class signaled by the initial pixel seed. The grown regions therefore constitute suitable training sets for each class. Comparing the statistical behavior of the pixel population of a sliding window with that of each class performs the classification. For region-growth, a window size is employed for each class. For classification, the centered pixel of the sliding window is labeled as belonging to a class if its spectral distance is a minimum to the class. The window size used for classification is a function of the best separability between the classes. A series of examples, employing synthetic and satellite images are presented to show the value of this classifier. The goodness of the segmentation is evaluated by means of the k coefficient and a visual inspection of the results.

A divergence operator to quantify texture from multi-spectral satellite images

A divergence operator to measure the texture content in a multi‐spectral image is proposed. A multi‐spectral image is modelled as an n‐dimensional vector field, ‘n’ being the number of bands of the image. A pixel of the image is an n‐dimensional vector in this field. It is demonstrated that the flux variations of the vector field are related to changes in texture in the image. A divergence operator measures the flux variation and, hence, texture. In order to save computer memory, speed up the divergence operator calculation and lessen the content of noise of the image, principal component transformation is applied to the bands of the image. The first three principal components are used to span the vector field. The partial derivatives involved in the divergence operator are written as weighted finite differences. To estimate these derivatives, cubes of three, five and seven voxels per side are considered. The cube is systematically displaced to cover the entire domain of the vector field. In each position of the cube, the divergence value is calculated using the weighted finite difference approximation. This value is written as a pixel in an output image file according to the Cartesian coordinates defined by the location of the cube. This image file depicts the texture variations of the multi‐spectral image. The relation flux variation versus coarseness of texture is discussed. Two examples, based on Landsat Thematic Mapper multi‐spectral and synthetic multi‐spectral images, are presented and analysed.

Edge enhancement in multispectral satellite images by means of vector operators

Resumen Edge enhancement is an element of analysis to derive the spatial structure of satellite images. Two methods to extract edges from multispectral satellite images are presented. A multispectral image is modeled as a vector field with a number of dimensions equal to the number of bands in the image. In this model, a pixel is defined as a vector formed by a number of elements equal to the number of bands. Two vector operators are applied to such vector field. In our first method, we extend the definition of the gradient. In this extension, the vector difference of the window central pixel with neighboring pixels is obtained. A multispectral image is then generated where each pixel represents the maximum change in spectral response in the image in any direction. This image is named a multispectral gradient. The other method, considers the generalization of the Laplacian by means of an h-dimensional Fourier transform. This image is named a multispectral Laplacian. The vector operators perform a simultaneous extraction of edgecontent in the spectral bands of a multispectral image. Our methods are parameter-free. Our methods work for a multispectral image of any number of bands. Two examples are discussed that involve multispectral satellite images at two scales. We compare our results with widely used edge enhancement procedures. The evaluation of results shows better performance of proposed methods when compared to widely used edge operators. Palabras clave: edge detection, multispectral image, edge enhancement, vector operator.

Detection of Maya's archaeological sites using high resolution radar images

The Yucatan peninsula, in the Mexican republic, harbours many archaeological sites of the Maya civilization. Many of these sites are covered by dense vegetation in areas of difficult access and high concentration of clouds all year round. Under these conditions, radar images present an option for archaeological prospecting. In this research, an area in the eastern part of the Yucatan peninsula has been selected where many Maya sites are located. For such an area, a radar image of the Radarsat system was acquired. Based on mathematical morphology, the speckle in this image was reduced using a geometric filter. Once the reduction of speckle is achieved, the detection of archaeological sites is accomplished by means of a series of grey scale morphological transformations applied to the image. The aforementioned transformations are designed to suppress the clutter of the vegetation and to enhance the archaeological sites at the same time. The existence of archaeological sites detected by our methodology is confirmed with ancillary data collected in the field and published literature.

A Model of Desertification Process in a Semi-Arid Environment Employing Multi-Spectral Images

A model of desertification in semi-arid environment employing satellite multi-spectral images is presented. The variables proposed to characterize desertification are: texture of terrain, vegetation index for semi-arid terrain, and albedo of terrain. The texture is derived from a divergence operator applied upon the vector field formed by the first three principal components of the image. The vegetation index selected is the TSAVI, suitable for semi-arid environment where vegetation is scarce. The albedo is calculated from the first principal component obtained from the bands of the multi-spectral image. These three variables are input into a clustering algorithm resulting in six desertification grades. These grades are ordered from no-desertification to severe desertification. Details are provided for the computer calculation of the desertification variables, and the parameters employed in the clustering algorithm. A multi-spectral Landsat TM image is selected for this research. A thematic map of desertification is then generated with the support of ancillary data related to the study area.

A method to derive texture-relief from ASTER bands 3N and 3B

The Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) onboard the TERRA satellite generates a pair of bands from which a stereoscopic view can be derived. These bands, called 3N and 3B, correspond to the spectral range 0.78–0.86 μm. In this work, a method based on principal component analysis (PCA) was used to derive, from these bands, the texture-relief of the terrain. We compare our method with a digital elevation model (DEM) obtained from an algorithm that takes advantage of the stereoscopic view. Two examples are presented and discussed in detail.

Characterization of vegetation in the south of Mexico by means of a canonical expansion

In this work, a method to characterize vegetation cover types is presented. The state of vegetation is represented by means of canonical variables derived from the bands of a multi-spectral image. The canonical variables handled in this work are the following: albedo of vegetation, strength of the greenness of vegetation, and humidity content of vegetation. These variables are uncorrelated, as is demonstrated in this research. The ensemble of these variables establishes a canonical expansion of the multi-spectral image. This is a canonical representation of vegetation. The canonical variables mentioned above form a multi-band image. This image is input into an unsupervised spectral classifier to generate a thematic map of vegetation cover types. Based on ground data, these vegetation classes are identified. In addition to this map, the following information layers are incorporated into the characterization process: lithological units, digital terrain model, hydrography, geomorphological units and climate. The whole of this information is integrated into a geospatial database to provide a description of the vegetation classes.

Morphological Change of a Scene Employing Synthetic Multispectral and Panchromatic Images

Climate change has produced transformations in the coastal zone of Tamaulipas State. Such changes include modifications to coastline and transformations to texture-relief and texture of the zone. In this work, high resolution panchromatic SPOT images have been employed to quantify such modifications. A synthetic multispectral image is used to validate our results. To quantify the texture-relief and texture, the multi-spectral image is modeled as a vector field of as many dimensions as bands of the image. Upon this field, the vector operators divergence and laplacian are applied. Results are presented for an area of Tampico-Altamira, details of the methodology are shown and results are discussed.

Multivariate classification of landscape metrics in multispectral digital images

Abstract. The use of landscape metrics to characterize the morphological behavior of a landscape has been extensive in the last few years. It is recognized that a single metric is insufficient to characterize a landscape. Such metrics are used individually to derive the morphological aspect of a landscape. No joint use of various metrics has been reported. Therefore, we considered the joint use of landscape metrics in a multivariate classification. We derived the value of a number of landscape metrics of patches from several case studies. A multivariate classification was applied using a hierarchical clustering algorithm. The multivariate classification was carried out using the least correlated landscape metrics. To consider the multivariate classification, a normalization of metrics range was used. The results provided the morphological structure of patches grouped into four or five classes. The classes depicted a morphological structure of patches that ranged from simple to very complex. An index was proposed to quantify the morphological structure of a class-patch. Such an index was defined as the average of the landscape metrics for a class-patch. The distance among the class-patch was given by means of the Jeffries–Matusita distance.