Vinciane Lacroix

A three-module strategy for edge detection

The first module is a parallel process computing local edge strength and direction, while the last module is sequential process following edges. The originality of the overall method resides in the intermediate module, which is seen as a generalization of the nonmaximum-deletion algorithm. The role of this module is twofold: It enables one to postpone some deletion to the last module where contextual information is available, and it transmits the local edge direction in order to guide the contour following. A postprocessing method called learning edges is proposed as a refinement of the method. The binary edge images extracted from various gray-level images illustrate the power of the strategy. >

Feature extraction using the constrained gradient

Abstract Low-level operators are needed in most computer vision applications in order to get relevant image primitives. In this paper, we present a line and a corner detector. Both operators use specific constraints on the gradient of the image intensity. The operators are applied to satellite and aerial images. The line detector is very useful for extracting roads even on the noisy SAR images, while the corner detector enables to detect salient points such as corners of buildings in aerial images. The information brought by these detectors completes the edge-based description of an image.

The primary raster: a multiresolution image description

The primary raster is a multiresolution image description. This description has the form of a raster where each pixel (edgel) has four characteristics: the detection scale, the blurring scale, the local contrast, and an edgel type. The detection scale is the finest resolution where the edgel appears, while the blurring scale is the coarsest resolution where the edgel is still present. The local contrast is the difference of the mean intensities taken from each side of the edgel. The edgel type depends on the evolution of the gradient from the finest to the coarsest resolution. Experimental results on the primary raster are presented. It is concluded that the primary raster provides a basis for a higher-level image description.<<ETX>>

Edge and line detection in polarimetric SAR images

A scheme for detecting edges and lines in multichannel SAR images is proposed. The line detector is constructed from the edge detector. The latter is based on multivariate statistical hypothesis tests applied to log-intensity SAR images. The raw results are vectorized by a traditional bright line extraction process. The scheme is illustrated by extracting dark linear structures on various full-polarimetric SAR images.

Detecting urbanization changes using SPOT5

An automatic system to estimate the urbanization changes on the Belgian territory, using SPOT5 images and the National Geographic Institute vectorial database is proposed. The images and the vectorial data are first co-registered. Then, the vectorial database is projected and dilated to produce a mask representing the old status of the database. On the other hand, a fusion of two classification processes on the images enables to extract the built-up area and the communication network, providing a mask representing the actual state of the urbanization in the zone. The comparison between the two masks gives a coarse information of the changes.

Edge detection: what about rotation invariance?

In this article, we consider techniques that provide edge strength at each pixel using two orthogonal directions and question their invariance with respect to a rotation.

Median Graph Shift: A New Clustering Algorithm for Graph Domain

In the context of unsupervised clustering, a new algorithm for the domain of graphs is introduced. In this paper, the key idea is to adapt the mean-shift clustering and its variants proposed for the domain of feature vectors to graph clustering. These algorithms have been applied successfully in image analysis and computer vision domains. The proposed algorithm works in an iterative manner by shifting each graph towards the median graph in a neighborhood. Both the set median graph and the generalized median graph are tested for the shifting procedure. In the experiment part, a set of cluster validation indices are used to evaluate our clustering algorithm and a comparison with the well-known Kmeans algorithm is provided.

Pixel labeling in a second-order markov mesh

Abstract In this paper, we address the pixle labeling problem under the assumption that contextual information in the picture is encoded in the transition probabilities of a second-order Markov mesh random field. We show that the problem can be solved by a linear algorithm generalizing Devijvers F-G-H algorithm. In order to label the current pixel, all the dependencies inside a triangular past neighborhood are taken into account. However, the computation of the probability of the labels lying at the border of the neighborhood is based upon an hypothesis. The complexity of the algorithm increases with the size of the considered neighborhood. Some experiments illustrating the theory proposed show how to recover an artificial binary image from its noisy version and how the size of the considered neighborhood influences the recovering. The very good results justify the use of this algorithm and moreover show that a best compromise between computer-time and precision leads to the F-G-H algorithm.

A multiresolution-MRF approach for stereo dense disparity estimation

We present a method that reduces the computational cost of the MRF-based stereo algorithm and increases the quality of the final disparity map. In a first step, using window-based method we compute successive disparity maps at different resolutions by varying the correlation window size, in order to estimate for each pixel the set of most probable disparity values. Thus, by replacing the initial disparity range — which may exceed hundreds of pixels for some applications — by the small set of valid disparities, we increase the probability of choosing the right value for each pixel and thus speed up the MRF optimization process.

Raster-to-Vector Conversion: Problems and Tools Towards a Solution A Map Segmentation Application

After the analysis of the problems a raster-to-vector (R2V) software can meet, a strategy involving a pre-processing, a clustering, a labeling and finally a vectorization phase is proposed. Much emphasis is put on the clustering and the labeling phases which depend on the pixel type (edge, line, or other). In particular, it is suggested to use the median-shift on all the pixels but the edgels to extract the main colors. Results are shown on scanned maps.