Carmen Patrascu

A Latent Analysis of Earth Surface Dynamic Evolution Using Change Map Time Series

With a continuous increase in the number of Earth Observation satellites, leading to the development of satellite image time series (SITS), the number of algorithms for land cover analysis and monitoring has greatly expanded. This paper offers a new perspective in dynamic classification for SITS. Four similarity measures (correlation coefficient, Kullback-Leibler divergence, conditional information, and normalized compression distance) based on consecutive image pairs from the data are employed. These measures employ linear dependences, statistical measures, and spatial relationships to compute radiometric, spectral, and texture changes that offer a description for the multitemporal behavior of the SITS. During this process, the original SITS is converted to a change map time series (CMTS), which removes the static information from the data set. The CMTS is analyzed using a latent Dirichlet allocation (LDA) model capable of discovering classes with semantic meaning based on the latent information hidden in the scene. This statistical method was originally used for text classification, thus requiring a word, document, corpus analogy with the elements inside the image. The experimental results were computed using 11 Landsat images over the city of Bucharest and surrounding areas. The LDA model enables us to discover a wide range of scene evolution classes based on the various dynamic behaviors of the land cover. The results are compared with the Corinne Land Cover map. However, this is not a validation method but one that adds static knowledge about the general usage of the analyzed area. In order to help the interpretation of the results, we use several studies on forms of relief, weather forecast, and very high resolution images that can explain the wide range of structures responsible for influencing the dynamic inside the resolution cell.

Normalized Autobinomial Markov Channels For Pedestrian Detection.

Pedestrian detection represents one of the most important components of engineering devices that use automated vision to help decision systems take quick and accurate actions. Such systems are defined and customized to be useful for different needs, such as monitoring and aided surveillance, or increasing safety features in automotive industry. Given the large spectrum of applications that use pedestrian detection, demand has increased in recent years for the development of feasible solutions which can be integrated in devices such as smartphones or action cameras. This paper focuses on finding probabilistic features that highlight the human body characteristics regardless of contextual information in images. Adjacent pixels are often spatially correlated, which means that they are likely to have similar values. We view the image as a collection of random variables indexed by certain locations, called sites. The state of a site ξ is conditionally independent of all variables in the random field, except the neighbouring system Nξ = { η ∈Ω | η 6= ξ , d2(ξ ,η)≤ ∆ } , where ∆ is a positive integer and d2(ξ ,η) is the squared Euclidean distance between ξ and η . The neighbouring system strictly depends on a collection of cliques C = ∑ ) k=1 Ck, where ω(∆) is the number of cliques for each local specification. Energy function: An unpublished manuscript [2] describes how to interpret the local property of a Markov random field in terms of energy and potential, claiming that the probability at a site ξ is given by:

Robust detection and tracking of salient face features in color video frames

A set of methods for automatic detection and tracking of human skin and of salient facial features, namely lips and eyes, based on the combination of color segmentation and shape extraction is presented. The algorithm does not require a training procedure or parameter set. The main addressed challenge is to provide efficient computation so to support real-time applications. This was achieved by performing all the computation in the RGB color space, by replacing full-detection on most of the frames with faster tracking and by reducing the features search area.

Performance testing and functional limitations of Normalized Autobinomial Markov Channels

The field of pedestrian detection has gained momentum in recent years, due to a large range of applications, including advanced robotics, aided surveillance and automotive safety. Its importance in the field of computer vision is confirmed by the large number of available algorithms, as well as the increased complexity of the public databases used for testing. To comply with the increased demands of the field, we perform extensive performance testing of the proposed Normalized Autobinomial Markov Channels (NAMC) algorithm using the Caltech Pedestrian Dataset. The proposed solution aims at isolating easily distinguishable body characteristics by learning contextual probabilistic dependencies. The functional limitations of the algorithm are derived by separately analyzing three test scenarios: scale, occlusion, aspect ratio. The obtained results demonstrate the efficiency of our approach, especially in the case of heavy occlusions, where the algorithm ranks first among the tested state-of-the art solutions.

Damage Assessment Based on SAR Image Analysis: Flood Scenario for Romanian Eastern Carpathian Region

When natural disasters occur, it is necessary for the authorities to make fast and effective decisions in order to prevent the occurrence of more damage, as well as to find solutions for the affected population that needs to be relocated. Satellite imagery can prove to be a useful instrument in decision support during emergency situations of such nature (floods), and especially SAR data, due to its all weather capabilities. This paper makes an assessment of the utility of satellite radar products (TerraSAR-X and Radarsat) in the frame of emergency situations management. A real case study is presented, where radar data were processed by human specialists on one hand, and automatically on the other hand, using an intelligent information extraction system.

Class evolution data analytics from sar image time series using information theory measures

In this paper we present the result of data analytics techniques applied to a database comprising of 32 SLC SM TerraSAR-X images, acquired over the area of Bucharest, Romania. The methodology follows a two step approach. The first stage consists of a coarse identification of potentially changed areas using a supervised learning image annotation tool with relevance feedback. Gabor texture features are used to describe image patches. The patch size is derived as a function of the resolution and pixel spacing of the data. In the second stage we apply an information theory strategy to refine the regions previously shown to exhibit class dynamics within the image stack, with pixel accuracy. Finally, a series of analytical indicators (absolute extent of areas affected by change, class evolution trends, inter-class correlations) are derived, in order to generate a predictive model for the selected test site.

Semantic annotation of high resolution TerraSAR-X images using Information Mining

This paper addresses the problem of High Resolution Synthetic Aperture Radar (SAR) image semantic annotation using a Knowledge Based Information Mining (KIM) System. The authors propose the assessment of the capabilities of KIM to perform an automatic urban classification on TerraSAR-X data. Four test sites have been used in the experiment to prove that the system is generic and data independent. The performance is evaluated by matching the results with GeoEye optical representations of the selected areas. For the evaluation a number of three classes are presented and discussed (water bodies, green urban areas and tall buildings).

Improving public datasets markings' quality using unsupervised refinement kernels

In recent years there has been an exponential growth in developing machine learning algorithms, focused on applications ranging from scene understanding, to the more standard object recognition and classification tasks. Although multiple approaches have been proposed for solving these issues, a common prerequisite is the existence of large datasets, which can be used both for training and testing purposes. We propose a semi-automatic annotation framework for object instances, which addresses the problems related to the big data paradigm in the context of object detection and pixel-level segmentation. The designed marking and learning workflow aims to be a cyclical process allowing iterative improvements of the marking architecture. Results of this processing chain are empirically validated on the COCO database.

AutoMarkov DNNs for object classification

Recent advances in the area of Deep Convolutional Neural Networks have led to steady progress, mainly observed in the field of object classification and localization. Extensive testing helped generate frameworks guaranteeing the initiation of successful network architectures. For this reason, the authors focus on bringing added value on specific nodes of a generic network configuration. We propose a novel type of convolutional layer based on Autobinomial Markov-Gibbs Random Fields (AutoMarkov Layer). Our choice is motivated by the fact that each neuron in a layer is only connected to a local region in the following layer. This property allows us to integrate Markov Random Fields into the structure of a neuron, to account for the probability of each particular pathway. Functional testing is performed on the MNIST, CIFAR-10 and CIFAR-100 datasets, showing clear improvements for correct classification scores on all the datasets mentioned regardless of the network architecture.

Multi-set analysis Of platform influence on density and quality of Persistent Scatterers: A case study for Bucharest area

During the past years, SAR techniques like Persistent Scatterer Interferometry (PSInSAR) have provided hyper-precision sensing at very large spatial scales. The continuous improvement in the quality of PS measurements comes from the constant development of new acquisition geometries embedded in various platforms. In this study we make a comparative assessment of the quality, number and density of Persistent Scatterers obtained using data acquired in different geometric configurations implemented on three platforms - ERS, ENVISAT, TerraSAR-X. All results were obtained by considering individual datasets of the same urban area (Bucharest), with a combined period of acquisitions of 22 years. The analysis is performed in terms of incidence angles, baseline, orbit type, look direction and PS dynamic range.