Hierarchical segmentation, object detection and classification in remotely sensed images

Abstract

Automatic content extraction and classification of remotely sensed images have become highly desired goals by the advances in satellite technology and computing power. The usual choice for the level of processing image data has been pixelbased analysis. However, spatial information is an important element to interpret the land cover because pixels alone do not give much information about image content. Automatic segmentation of high-resolution remote sensing imagery is an important problem in remote sensing applications because the resulting segmentations can provide valuable spatial and structural information that are complementary to pixel-based spectral information in classification. In this thesis, we first present a method that combines structural information extracted by morphological processing with spectral information summarized using principal components analysis to produce precise segmentations that are also robust to noise. First, principal components are computed from hyper-spectral data to obtain representative bands. Then, candidate regions are extracted by applying connected components analysis to the pixels selected according to their morphological profiles computed using opening and closing by reconstruction with increasing structuring element sizes. Next, these regions are represented using a tree, and the most meaningful ones are selected by optimizing a measure that consists of two factors: spectral homogeneity, which is calculated in terms of variances of spectral features, and neighborhood connectivity, which is calculated using sizes of connected components. Experiments on three data sets show that the method is able to detect structures in the image which are more precise and more meaningful than the structures detected by another approach that does not make strong use of neighborhood and spectral information.Then, we introduce an unsupervised method that combines both spectral and structural information for automatic object detection. First, a segmentation hierarchy is constructed and candidate segments for object detection are selected by the proposed segmentation method. Given the observation that different structures appear more clearly in different principal components, we present an algorithm that is based on probabilistic Latent Semantic Analysis (PLSA) for grouping the candidate segments belonging to multiple segmentations and multiple principal components. The segments are modeled using their spectral content and the PLSA algorithm builds object models by learning the objectconditional probability distributions. Labeling of a segment is done by computing the similarity of its spectral distribution to the distribution of object models using Kullback-Leibler divergence. Experiments on three data sets show that our method is able to automatically detect, group, and label segments belonging to the same object classes. Finally, we present an approach for classification of remotely sensed imagery using spatial information extracted from multi-scale segmentations. Different structuring element size ranges are used to obtain multiple representations of an image at different scales to capture different details inherently found in different structures. Then, pixels at each scale are grouped into contiguous regions using the proposed segmentation method. The resulting regions are modeled using the statistical summaries of their spectral properties. These models are used to cluster the regions by the proposed grouping method, and the cluster memberships assigned to each region at multiple scales are used to classify the corresponding pixels into land cover/land use categories. Final classification is done using decision tree classifiers. Experiments with three ground truth data sets show the effectiveness of the proposed approach over traditional techniques that do not make strong use of region-based spatial information.