An unsupervised algorithm for hyperspectral image segmentation based on the Gaussian mixture model

Acito, N.; Corsini, Giovanni; Diani, Marco

doi:10.1109/igarss.2003.1295256

Cited by 40 publications

(30 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…To be able to visually compare the 6 images, we mapped the colors taking the 64-scan GMM case as a reference since it has the highest k, and mapping to minimize the distance between the means (l k ) of the Gaussians from one model to the other. 6.89, 21.46 and 33.87. The number of classes is not strictly identical among the various datasets, but they are consistent: between 14 and 16 for the GMM models and between 12 and 13 for the cGMM models.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Unsupervised Segmentation of Spectral Images with a Spatialized Gaussian Mixture Model and Model Selection

Cohen¹,

Pennec

2014

Oil Gas Sci. Technol. – Rev. IFP Energies nouvelles

View full text Add to dashboard Cite

and available online here Cet article fait partie du dossier thématique ci-dessous publié dans la revue OGST, Vol. 69, n°2, et téléchargeable ici Re´sume´-Me´lange de Gaussiennes spatialise´et se´lection de mode`le pour la segmentation nonsupervise´e d'images spectrales -Dans cet article nous de´crivons un nouvel algorithme de segmentation non-supervise´e applicable aux images spectrales. Cet algorithme e´tend les techniques de classification non-supervise´e fonde´es sur les mode`les de me´lange de Gaussiennes, en y incorporant les informations spatiales : les spectres sont mode´lise´s par un me´lange de K classes, chacune avec une distribution Gaussienne, dont les proportions de me´lange de´pendent de la position. En imposant une structure constante par morceaux aux proportions de me´lange, nous construisons une proce´dure d'estimation, de type maximum de vraisemblance pe´nalise´e, qui optimise simultane´ment la partition ainsi que les autres parame`tres du mode`le, en particulier le nombre de classes. Nous fournissons une garantie the´orique pour cette estimation, meˆme quand la loi ge´ne´ratrice ne fait pas partie des mode`les envisage´s, et de´crivons une mise en oeuvre efficace. Finalement, nous appliquons cet algorithme a`un jeu de donne´es re´el.Abstract -Unsupervised Segmentation of Spectral Images with a Spatialized Gaussian Mixture Model and Model Selection -In this article, we describe a novel unsupervised spectral image segmentation algorithm. This algorithm extends the classical Gaussian Mixture Model-based unsupervised classification technique by incorporating a spatial flavor into the model: the spectra are modelized by a mixture of K classes, each with a Gaussian distribution, whose mixing proportions depend on the position. Using a piecewise constant structure for those mixing proportions, we are able to construct a penalized maximum likelihood procedure that estimates the optimal partition as well as all the other parameters, including the number of classes. We provide a theoretical guarantee for this estimation, even when the generating model is not within the tested set, and describe an efficient implementation. Finally, we conduct some numerical experiments of unsupervised segmentation from a real dataset.

show abstract

Section: Discussionmentioning

confidence: 99%

“…The second approach has exactly the inverse behavior. The first technique has been used, for instance, by Tarabalka et al [4] and Bunte et al [5] while the second has been used by Acito et al [6] and Yang et al [7].…”

Section: Introductionmentioning

confidence: 99%

Unsupervised Segmentation of Spectral Images with a Spatialized Gaussian Mixture Model and Model Selection

Cohen¹,

Pennec

2014

Oil Gas Sci. Technol. – Rev. IFP Energies nouvelles

View full text Add to dashboard Cite

show abstract

“…Hyperspectral image segmentation using a multicomponent hidden Markov chain model has been proposed in [11]. A statistical hyperspectral image segmentation approach based on Gaussian mixture models is presented in [12]. In [13], it has been proposed to append texture information via filter banks to increase hyperspectral image segmentation accuracy (SA).…”

Section: Introductionmentioning

confidence: 99%

Unsupervised Segmentation of Hyperspectral Images Using Modified Phase Correlation

Ertürk¹,

Ertürk

2006

IEEE Geosci. Remote Sensing Lett.

View full text Add to dashboard Cite

This letter presents hyperspectral image segmentation based on the phase-correlation measure of subsampled hyperspectral data, which is referred to as modified phase correlation. The hyperspectral spectrum of each pixel is initially subsampled to gain robustness against noise and spatial variability, and phase correlation is applied to determine spectral similarity. Similar and dissimilar pixels are decided according to the peak value of the phase correlation result to determine pixels that fall into the same segments. The approach can be regarded as a region-growing technique. The total number of segments is determined automatically according to the similarity threshold.

show abstract

“…The ground resolution is of about 3 m. We have selected the pixels belonging to two different classes: 1) class #1: Grass (369951 pixels) and 2) class #2: Bare Soil (23482 pixels). For each class, pixel selection has been performed by using the unsupervised segmentation algorithm proposed in [10]. The rgb image of the scene and the selected classes are shown in Fig.1.…”

Section: Resultsmentioning

confidence: 99%

“…To estimate those parameters we adopt an approach consisting in searching the "best fitting" between empirical and theoretical pdfs. The goodness of fit is evaluated by a suitable cost function ( ) Ω P J calculated on P selected points (percentile) of the two pdfs and the estimate Ω is obtained by minimizing that function: (10) Note that the cost function evaluates the relative square error between the logarithm of the empirical and the theoretical pdfs. The logarithmic transformation is applied in order to give the same weight to the body and to the tails of the distributions.…”

Section: B Elliptically Contoured Distribution (Ecd) Modelmentioning

confidence: 99%

Hyper-spectral data modelling by non-Gaussian statistical distributions

Acito¹,

Corsini²,

Diani³

IEEE International IEEE International IEEE International Geoscience and Remote Sensing Symposium, 2004. IGARSS '04. Proceedings

Self Cite

View full text Add to dashboard Cite

In this manuscript we investigate on the statistical modeling of hyper-spectral data. Accurately modeling real data is of paramount importance in the design of optimal classification or detection strategies and in evaluating their performances. In the work three non-Gaussian models are considered and their capability in characterizing the statistical behavior of real data is discussed with reference to a data set acquired by the Multispectral Infrared and Visible Imaging Spectrometer (MIVIS) sensor

show abstract

An unsupervised algorithm for hyperspectral image segmentation based on the Gaussian mixture model

Cited by 40 publications

References 4 publications

Unsupervised Segmentation of Spectral Images with a Spatialized Gaussian Mixture Model and Model Selection

Unsupervised Segmentation of Spectral Images with a Spatialized Gaussian Mixture Model and Model Selection

Unsupervised Segmentation of Hyperspectral Images Using Modified Phase Correlation

Hyper-spectral data modelling by non-Gaussian statistical distributions

Contact Info

Product

Resources

About