Search

Building detection has been a prominent area in the area of image classification. Most of the research effort is adapted to the specific application requirements and available datasets. Our dataset includes aerial orthophotos (with spatial resolution 20cm), a DSM generated from LiDAR (with spatial resolution 1m and elevation resolution 20 cm) and DTM (spatial resolution 2m) from an area of Athens, Greece. Our aim is to classify these data by means of Markov Random Fields (MRFs) in a Bayesian framework for building block extraction and perform a comparative analysis with other supervised classification techniques namely Feed Forward Neural Net (FFNN), Cascade-Correlation Neural Network (CCNN), Learning Vector Quantization (LVQ) and Support Vector Machines (SVM). We evaluated the performance of each method using a subset of the test area. We present the classified images, and statistical measures (confusion matrix, kappa coefficient and overall accuracy). Our results demonstrate that the MRFs and FFNN perform better than the other methods.

International audience; This paper investigates the use of a popular regularization model, the Total Variation minimization (TV), to filter SAR images and reduce speckle noise. This model is extensively used for its property of preserving edges. Due to the many local minima, TV minimization is difficult to achieve for non-convex likelihood terms such as that of SAR amplitude. Such a minimization can be performed efficiently by computing minimum cuts on weighted graphs. Exact minimization, although theoretically possible, can not be implemented due to memory constraints on large images required by remote sensing applications. The computational burden of the state-of-the-art algorithm for approximate minimization is also heavy. In this paper, we propose a new fast approximate discrete algorithm. The filtering is applied in the framework of building delineation for 3D reconstruction. Results on real images are presented.

We present a “continuous” analysis of a solar H-alpha image in order to address the problem of image segmentation. Our approach is based on combinatorial optimization methods and in particular on Mean Field Fast Annealing (MFFA). Mean-field theory gives a deterministic nature to our algorithm while its efficiency is improved by a fast cooling schedule. We show how this method call be used to separate efficiently the regions of different solar activity giving a tool for a future automated recognition and classification of sunspots.

We demonstrate a close relationship between classical models from statistical physics and Markov randomfield models for image labeling purposes. A picture is taken of a real image or map, defined by a measure of intensity over a set of pixels. Possible class (or gray-level) values are assigned to spin values in Potts theory. We present a continuous analysis of image relaxation by mean-field theory and apply it by using standard and extended Potts models. Accurate relaxation results were obtained with a specific deterministic method called mean-field fast annealing. © 1997 Optical Society of America.