Jhung, Y., and Swain, P. H. 1996. Bayesian Contextual Classification Based on Modified M-Estimates and Markov Random Fields. IEEE Transaction on Pattern Analysis and Machine Intelligence 34(1):67--75.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....are independent of the properties of all other pixels. Consequently, the MLP classifier may have difficulty distinguishing the pixels that come from different land cover classes but have very similar spectral properties. The result is often a snow like classification map. Several studies [1] [2] have shown that a contextual classifier that utilizes both spectral and spatial contextual information are able to better discriminate between the pixels with similar spectral attributes but located in different regions, allow reduction of the speckle error, and improve the classification ....

....allow reduction of the speckle error, and improve the classification performance significantly. However, this type of classifier also faces the problem of the small training sample size where the class conditional probability density must be estimated in the analysis of hyperspectral data [1][2]. In [3] it has been demonstrated that an adaptive maximum likelihood pixel classifier (AMLP) may alleviate the small training sample problem by including semi labeled samples along with the training samples during the process of statistics estimation. Essentially, this classifier is formed by ....

Yonhong Jhung and Philip H. Swain, " Bayesian contextual classification based on modified M-estimates and markov random fields", IEEE Trans. Geosci. Remote Sensing, vol.34, no. 1, pp. 68-75, Jan. 1996

....ABC ML classifier instead of the ABC MAP classifier as a conventional ICM does. The reason for the third modification is as follows. First, it has been shown that in the ICM starting with the classification results from a ML classifier, in general the MAP classifier outperforms the ML classifier [9, 10]. Even though a postprocessing classifier may be able to improve classification accuracy also by reducing the speckle error, it is more likely to be overdone and lead to loss of more details than using the ABC MAP classifier. In other words, semi labeled samples generated from the ABC MAP ....

Yonhong Jhung and Philip H. Swain, Bayesian contextual classification based on modified M-estimates and markov random fields , IEEE Trans. Geosci. Remote Sensing, vol.34, no. 1, pp. 68-75, Jan. 1996.

....0 0 1 1 1 0 0 A B C D A B C D 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0 0 0 0 (b) Neighbor relationship (c) Contiguity Matrix 1 Figure 3.3: A spatial framework and its four neighborhood contiguity matrix. Modeling Spatial Dependencies Using the SAR and MRF Models Several previous studies [Jhung Swain1996], Solberg, Taxt, Jain1996] have shown that the modeling of spatial dependency (often called context) during the classification process improves overall classification accuracy. Spatial context can be defined by the relationships between spatially adjacent pixels in a small neighborhood. The ....

Jhung, Y., and Swain, P. H. 1996. Bayesian Contextual Classification Based on Modified M-Estimates and Markov Random Fields. IEEE Transaction on Pattern Analysis and Machine Intelligence 34(1):67--75.

....pepper noise. These classifiers also suffer in terms of classification accuracy. There are two major approaches for incorporating spatial dependence into classification prediction models: spatial autoregression models [2] 15] 16] 17] 23] 24] and Markov Random Field models [5] 6] 9] [13] [18] 29] 31] Here we want to make a note regarding the terms spa tial dependence and spatial context. These words originated in two different communities. Natural resource analysts and statisticians use spatial dependence to refer to spatial autocorrelation and the image processing community ....

....to provide better models than logistic regression in terms of achieving higher confidence (R2) Similarly, Markov Random Fields (MRFs) is a popular model for incorporating spatial context into image segmentation and land use classification problems. Over the last decade, several researchers [29] [13], 31] have exploited spatial context in classification using Markov Random Fields to obtain higher accuracies over their counterparts (i.e. non contextual classifiers) MRFs provide a uniform framework for integrating spatial context and deriving the probability distribution of interacting ....

[Article contains additional citation context not shown here]

Yonhong Jhung and Philip H. Swain. Bayesian Contextual Classification Based on Modified M-Estimates and Markov Random Fields. IEEE Transaction on Pattern Analysis and Machine Intelligence, 34(1):67 75, 1996.

....namely the Spatial Autoregressive Model (SAR) and Markov Random Fields (MRF) and analyze their performance in an example case, the prediction of the location of bird nests in the Darr and Stubble wetlands. 2. 1 Modeling Spatial Dependencies Using the SAR and MRF Models Several previous studies [13], 30] have shown that the modeling of spatial dependency (often called context) during the classification process improves overall classification accuracy. Spatial context can be defined by the relationships between spatially adjacent pixels in a small neighborhood. The spatial relationship among ....

Yonhong Jhung and Philip H. Swain. Bayesian Contextual Classification Based on Modified M-Estimates and Markov Random Fields. IEEE Transaction on Pattern Analysis and Machine Intelligence, 34(1):67-75, 1996.

....statistics, and each cycle is started with a ML classifier instead of a MAP classifier. The reason for this choice is as follows. First, it has been shown that in ICM starting with the classification results from a ML classifier, in general the MAP classifier outperforms the ML classifier [33] [34]. Even though a postprocessing process may be able to improve classification accuracy also by reducing the speckle error, it is more likely to be overdone and lead to loss of details. Therefore, semi labeled samples generated from the MAP classifier should contain more correctly classified ....

Yonhong Jhung and Philip H. Swain, Bayesian contextual classification based on modified M-estimates and markov random fields, IEEE Trans. Geosci. Remote Sensing, vol.34, no. 1, pp. 68-75, Jan. 1996

....and only by the classes assigned to the pixels x 0 in its neighborhood N (x) 4] Here we define N as the quadrangular k Theta k window around the pixel x, excluding x itself. MRF based approaches to contextually enhanced multispectral classification have recently been used in remote sensing [3,7]. Unlike these approaches we here use a contextual potential function U(xj ) which not only evaluates identical or different labels in the neighborhood N with a f0,1g Kronecker function, but which is influenced continuously by the current probabilities of the neighboring pixels: p con (xj ) ....

....= X x 0 2N (x) 1 Gamma P ( jx 0 ) 4) Then the context conditional probability p con (xj ) is computed from the neighborhood potential U(xj ) where the parameter fi defines the magnitude of the contextual influence. For fi = 0 the influence is vanishing, and fi = 1 is a common choice [1,3,7]. 3 The Iterative Algorithm The probabilities and the necessary parameters are estimated during an iterative process using the current conditional probabilities (similar to the recently often used ICM algorithm (iterated conditional mode) 1] In our case the conditional probabilities depend ....

Y. Jhung and Philip H. Swain. Bayesian contextual classification based on modified M- estimates and Markov random fields. IEEE T.o.Geosci.a.Rem.Sens., 34(1):67--75, 1996.

No context found.

Jhung, Y., and Swain, P. H. 1996. Bayesian Contextual Classification Based on Modified M-Estimates and Markov Random Fields. IEEE Transaction on Pattern Analysis and Machine Intelligence 34(1):67--75.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC