Time-Varying Segmentation for Mapping of Land Cover Changes

Among all satellite imaging systems, Synthetic Aperture Radar (SAR) is useful for monitoring purposes, as they provide data at any time under all weather conditions. Today, three spaceborne SAR systems are operational, while by the end of 2007 three further SAR instruments will be launched, thereby allowing to obtain an almost continuous monitoring of the Earth coverage. In order to translate these multi-temporal data into information in an unsupervised (automated) and reliable way, sophisticated algorithms must be available. In the past years several approaches - primarily based on texture analysis and statistical scene estimation - have been proposed for multi-temporal SAR data analysis. Such methods, based on probability density functions, perform well under strictly controlled conditions, but they are often limited with respect to sensor synergy where complex joint probability density functions must be considered - and to the temporal aspect. To address these limitations, an original set of algorithms for the unsupervised multi-temporal SAR data analysis is proposed. To this end, several issues have been tackled: filtering, edge detection, and image segmentation. Moreover, condition sine qua non for the system design, was that i) it is sensor independent, and ii) data from different SAR systems can be ingested without any a-priori knowledge about the probability density function. Basically, the proposed time varying segmentation involves four independent steps. In a first step, a multi-temporal anisotropic non-linear diffusion filter is applied to filter the images which ultimately allow feature extraction. Subsequently, an extension of Canny edge detection algorithm for multi-temporal edge detection is applied, hence obtaining an edge map consistent across the whole sequence of temporal images. In a third step, closed regions are obtained using a two-part coding scheme with the edge map (as side information) and region growing technique (using the multi-temporal stack). Finally, the number of underlying spectral class composing a segment histogram at every frame is estimated, thus detecting changes due to temporal land cover fluctuations. Results are presented based on a set of 16 ENVISAT ASAR Alternating Polarization and Radarsat-1 Fine Beam images acquired between November 2004 and July 2005 for an agricultural (maize and sun flower) area in South Africa.

Exploiting Multi-Temporal Information for SAR Image Segmentation

François Aspert, Meritxell Bach Cuadra, Jean-Philippe Thiran

Earth observation is taking more and more importance in our society. Indeed, recent technologies march allows for satellite images to get ground resolution up to less than one meter. Thus, most of satellite image information obtained begins now to be widely used for multiple purposes ranging from cartography and mapping to agricultural management and security applications. Among all the satellite imaging systems, Synthetic Aperture Radar (or SAR) are very interesting for global earth surveillance and monitoring purposes as they can provide data at any time under all weather conditions. However only few work have been done to take advantage of such time robustness properties for retrieving the images main structures. In this study we focused one the particular issue of obtaining an efficient closed region segmentation using multi-temporal SAR image series. To this end, several tools have been developed. In order to remove SAR inherent noise we filtered the image sequences using a multi-temporal anisotropic non-linear diffusion algorithm. Then a new approach of Canny edge detection have been introduced to extract edge features from these vector-valued images. Finally, a region segmentation step has been applied to get closed regions. We demonstrated that for all the methods, exploiting multi-temporal information can improve the results accuracy. Moreover, since some radiometric changes can occur within region due to man intervention or natural diseases we also presented a direct application of region segmentation to retrieve such changes across time series.

2006

Automated Detection and Delineation of Tree Structures in Noisy Micrographs

Germán González Serrano

Despite more than a century of research, we still lack a concise description of the different types of neurons in the brain. This is a result of their large variance in morphology and the time consuming reconstruction process. In this thesis we design novel fully automated algorithms to release neuroscientists from such burdern. To that end, we first propose a filtering method that enhances neurite structures while rejecting structured noise. Prior work rely on filters optimized for signals of a particular shape, such as an ideal edge or ridge. While these approaches are optimal when the image conforms to these ideal shapes, their performance quickly degrades on many types of real data where the image deviates from such ideal model. We show that by learning rotational features, we can outperform state-of-the-art filament detection techniques on many different kinds of imagery. More specifically, we demonstrate superior performance for the detection of blood vessel in retinal scans, neurons in brightfield microscopy imagery, and streets in satellite imagery. Afterwards we propose algorithms that extract automatically the tree structures present on the images by maximizing a global likelihood under a generative model in noisy 2D images and 3D image stacks. Unlike earlier methods that rely mostly on local evidence, our method builds a set of candidate trees over many different subsets of points likely to belong to the final one and then chooses the best one according to a global objective function. Since we are not systematically trying to span all nodes, our algorithm is able to eliminate noise while retaining the right tree structure. We performed extended evaluation of the proposed algorithms and participated on a prestigious neuron reconstruction challenge. We obtained an award "for our deeper potential, more original approach, and ultimate scalability of our proposed solution". Finally, we propose system that tracks, segments and traces neurons in time-lapse microscopy and is suitable for high throughput image analysis. We examined neuronal differentiation datasets where gene expressions were inhibited using RNA interference. We were able to corroborate previously known behaviours and to derive new ones.

EPFL2011

Transformation-Invariant Analysis of Visual Signals with Parametric Models

Elif Vural

The analysis of collections of visual data, e.g., their classification, modeling and clustering, has become a problem of high importance in a variety of applications. Meanwhile, image data captured in uncontrolled environments by arbitrary users is very likely to be exposed to geometric transformations. Therefore, efficient methods are needed for analyzing high-dimensional visual data sets that can cope with geometric transformations of the visual content of interest. In this thesis, we study parametric models for transformation-invariant analysis of geometrically transformed image data, which provide low-dimensional image representations that capture relevant information efficiently. We focus on transformation manifolds, which are image sets created by parametrizable geometric transformations of a reference image model. Transformation manifolds provide a geometric interpretation of several image analysis problems. In particular, image registration corresponds to the computation of the projection of the target image onto the transformation manifold of the reference image. Similarly, in classification, the class label of a query image can be estimated in a transformation-invariant way by comparing its distance to transformation manifolds that represent different image classes. In this thesis, we explore several problems related to the registration, modeling, and classification of images with transformation manifolds. First, we address the problem of sampling transformation manifolds of known parameterization, where we focus on the target applications of image registration and classification in the sampling. We first propose an iterative algorithm for sampling a manifold such that the selected set of samples gives an accurate estimate of the distance of a query image to the manifold. We then extend this method to a classification setting with several transformation manifolds representing different image classes. We develop an algorithm to jointly sample multiple transformation manifolds such that the class label of query images can be estimated accurately by comparing their distances to the class-representative manifold samples. The proposed methods outperform baseline sampling schemes in image registration and classification. Next, we study the problem of learning transformation manifolds that are good models of a given set of geometrically transformed image data. We first learn a representative pattern whose transformation manifold fits well the input images and then generalize the problem to a supervised classification setting, where we jointly learn multiple class-representative pattern transformation manifolds from training images with known class labels. The proposed manifold learning methods exploit the information of the type of the geometric transformation in the data to compute an accurate data model, which is ignored in previous manifold learning algorithms. Finally, we focus on the usage of transformation manifolds in multiscale image registration. We consider two different methods in image registration, namely, the tangent distance method and the minimization of the image intensity difference with gradient descent. We present a multiscale performance analysis of these methods. We derive upper bounds for the alignment errors yielded by the two methods and analyze the variations of these bounds with noise and low-pass filtering, which is useful for gaining an understanding of the performance of these methods in image registration. To the best of our knowledge, these are the first such studies in multiscale registration settings. Geometrically transformed image sets have a particular structure, and classical image analysis methods do not always suit well for the treatment of such data. This thesis is motivated by this observation and proposes new techniques and insights for handling geometric transformations in image analysis and processing.