Saturday, April 20th 2024, 09:41

Avik Bhattacharya

Document Actions

Former PhD Student, CNES / INRIA / ENST

Contact :

Mail :		AvikdotBhattacharyaatinriadotfr
Phone :		(33)4-92-38-75-68
Fax :		(33)4-92-38-76-43
Postal adress :		INRIA Sophia Antipolis 2004, route des Lucioles 06902 Sophia Antipolis Cedex France

Abstract :

My research is focused in the field of road network extraction and retrieval from remote sensing images. Although, many retrieval system appertain to remote sensing images, but here in this work I sought to explore the retrieval of road network pattern based on the technique of graph matching with an emphasis on remote sensing scenarios.

Short Bio :

'95-'98 - BSc (Bachelor of Science) in Mathematics, I.I.T Kharagpur, India.
'98-'00 - MSc (Master of Science) in Mathematics, I.I.T Kharagpur, India.
'00-'01 - Project Assistant, Dept. of Ocean Engg. and Naval Architecture, I.I.T Kharagpur, India.
'01-'04 - Research Assistant, Fraunhofer Institut für Techno-und Wirtschaftsmathematik (ITWM), Kaiserslautern, Germany.
'04-'' - Phd Student at ENST, Paris/INRIA, Sophia Antipolis, France.

Last publications in Ariana Research Group :

Indexing of mid-resolution satellite images with structural attributes.
A. Bhattacharya and M. Roux and H. Maitre and I. H. Jermyn and X. Descombes and J. Zerubia. In The International Society for Photogrammetry and Remote Sensing, Beijing, China, July 2008.
Keywords : Landscape, Segmentation, Features, Extraction, Classification, Modelling.

Abstract :

Indexing and retrieval of satellite images relies on the extraction of appropriate information from the data about the entity of interest
(e.g. land cover type) and on the robustness of this extraction to nuisance variables. Entities in an image may be strongly correlated
with each other and can therefore be used to characterize geographical environments on the Earth’s surface.
The properties of road networks vary considerably from one geographical environment to another. The networks pertaining in a
satellite image can therefore be used to classify and retrieve such environments. In the work presented in this paper we have defined
7 such classes. These classes can be categorized as follows: 2 urban classes consisting of “Urban USA” and “Urban Europe”; 3
rural classes consisting of “Villages”, “Mountains” and “Fields”; an “Airports” class and a “Common” class (this can be considered
as a rejection class). These classes were then classified with the aid of geometrical and topological features computed from the road
networks occurring in them. In our work we have used two extraction methods simultaneously on an image to extract the road networks
pertaining in it. A set of 16 network features were computed from one extraction method and were categorized into 6 groups as follows:
6 measures of ‘density’, 4 measures of ‘curviness’, 2 measures of ‘homogeneity’, 1 measure of ‘length’, 2 measures of ‘distribution’
and 1 measure of ‘entropy’.
Due to certain limitations of these extraction methods there was a relative failure of network extraction in certain urban regions con-
taining narrow and dense road structures. This loss of information was circumvented by segmenting the urban regions and computing
a second set of geometrical and topological features from them. A set of 4 urban region features were computed and were categorized
into 3 groups as follows: 2 measures of ‘density’, 1 measure of ‘labels’ and 1 measure of ‘compactness’.
The 500 images (each of size 512x512 pixels) forming our database were selected from SPOT5 scenes with 5m resolution. From each
image a set of geometrical and topological features were computed from the road networks and urban regions. These features were
then used to classify the pre-defined geographical classes. Feature selection was done to avoid the burden of feature dimensionality
and increase the classification performance. A set of 20 features was selected from 36 features by Fisher Linear Discriminant (FLD)
analysis which gave the least classification error with an one-vs-rest linear Support Vector Machine (SVM).
The impact of spatial resolution and size of images on the feature set have been explored in this work. We took a closer look at the effect
of spatial resolution and size of images on the discriminative power of the feature set to classify the images belonging to the pre-defined
geographical classes. Tests were performed with feature selection by FLD and one-vs-rest linear SVM classification on a database with
images of 10m resolution. Another test was performed with feature selection by FLD and one-vs-rest linear SVM classification on a
database with 5m resolution images (each of size 256x256 pixels).
With the above mentioned approaches, we developed a novel method to classify large satellite images acquired by SPOT5 satellite (5m
resolution) with patches of images each of size 512x512 pixels extracted from them. There has been a large amount of work dedicated
to the classification of large satellite images at pixel level rather than considering image patches of different sizes. Classification of
image patches of different sizes from a large satellite image is a novel idea in the sense that the patches considered contain significant
coverage of a particular type of geographical environment.
Road networks and urban region features were computed from these image patches extracted from the large image. A one-vs-rest
Gaussian kernel SVM classification method was used to classify this large image. The classification results show that the image
patches were labeled with the class having the maximum geographical coverage of the area associated in the large image. The large
image was mapped into a “region matrix”, where each element of the matrix corresponds to a geographical class. This is a ‘hard’
classification and no inference can be drawn about the classification confidence.
In certain cases, this produces some anomalies, as a single patch may contain two or more different geographical coverages. In order
to have an estimate of these partial coverages, the output of the SVM was mapped into probabilities. These probability measures were
then studied to have a closer look at the classification accuracies. The results confirm that our method is able to classify a large image
into various geographical classes with a mean error of less than 10%.
Future studies can use operators to detect not only man-made structures like roads and urban areas, but also natural entities like rivers,
forests, etc. In this work we have restricted ourselves to a single resolution, but our methodology can be adapted to consider images
of higher resolutions from QuickBird and the future Pleiade satellite. At a better resolution it may be possible to extract different
structures like buildings, gardens, cross-roads, etc. This in turn will allow us to incorporate more classes to appropriately classify any
geographical environment. At an image resolution of 1m, we may imagine to have sub-classes of an existing class, e.g., classes like
urban Europe and urban USA can de divided into downtown, residential and industrial classes.

Indexing Satellite Images with Features Computed from Man-Made Structures on the Earth’s Surface.
A. Bhattacharya and M. Roux and H. Maitre and I. H. Jermyn and X. Descombes and J. Zerubia. In Proc. International Workshop on Content-Based Multimedia Indexing, Bordeaux, France, June 2007.
Keywords : Indexation, Road network, Semantic, Retrieval, Feature statistics.

Indexing of satellite images using structural information.
A. Bhattacharya. PhD Thesis, Ecole Nationale Supérieure des Télécommunications, 2007.
Keywords : Landscape, Segmentation, Features, Extraction, Classification, Data mining.

All publications in Ariana Research Group

Navigation

Languages