Lloyd clustering of Gauss mixture models for image compression and classification

Aiyer, A.; Pyun, Kyungsuk; Huang, Ying-zong; O'Brien, Deirdre; Gray, Robert M.

doi:10.1016/j.image.2005.03.003

Cited by 44 publications

(27 citation statements)

References 55 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is paired with aired with its hand-labelled classified image. These photographs have also been studied by Aiyer et al (2005), Li et al (2000), Pyun et al (2007) and Pyun et al (2009). In the analysis, we To design computer-based system to measure the spatial correlation in land use, we first segment the images into manmade and natural area using a block-based classifier.…”

Section: Aerial Imagementioning

confidence: 89%

“…In the analysis, we use 4 × 4 sub-blocks (256 × 256 sub-blocks in total) and estimating the Gauss mixture model for manmade and natura regions using 5 training images. The block processing steps are same with those in Aiyer et al (2005) and Pyun et al (2007) and their details are omitted here. Using the estimate GMMs, we classify each sub-block into the class having the highest likelihood.…”

Section: Aerial Imagementioning

confidence: 99%

See 1 more Smart Citation

Hyper-Parameter in Hidden Markov Random Field

Lim

Yu²,

Pyu³

2011

Korean Journal of Applied Statistics

View full text Add to dashboard Cite

Hidden Markov random field(HMRF) is one of the most common model for image segmentation which is an important preprocessing in many imaging devices. The HMRF has unknown hyper-parameters on Markov random field to be estimated in segmenting testing images. However, in practice, due to computational complexity, it is often assumed to be a fixed constant. In this paper, we numerically show that the segmentation results very depending on the fixed hyper-parameter, and, if the parameter is misspecified, they further depend on the choice of the class-labelling algorithm. In contrast, the HMRF with estimated hyper-parameter provides consistent segmentation results regardless of the choice of class labelling and the estimation method.Thus, we recommend practitioners estimate the hyper-parameter even though it is computationally complex.

show abstract

Section: Aerial Imagementioning

confidence: 89%

Section: Aerial Imagementioning

confidence: 99%

Hyper-Parameter in Hidden Markov Random Field

Lim

Yu²,

Pyu³

2011

Korean Journal of Applied Statistics

View full text Add to dashboard Cite

show abstract

“…Therefore, it would be of interest to relax the nearest-neighbor constraint and allow more general encoders. For instance, one could use Gauss Mixture Vector Quantization (GMVQ) [1] to model the distribution of the features as a Gauss mixture, thus allowing for partition cells with quadratic boundaries. Going beyond information loss minimization, our approach readily extends to any Bregman divergence [6], [2], not just the relative entropy.…”

Section: Discussionmentioning

confidence: 99%

“…(2) can be interpreted as the difference between the amount of information provided by X about Y and the amount of information provided by K about Y , and minimizing it leads to a clustering that throws away as little information about Y as possible. 1 Unfortunately, the algorithm of [10] is not suitable for our target application of quantizer design. For one, quantization requires an encoding rule that works on continuous data, does not depend on the labels other than those of the training examples, and can be applied outside the training set.…”

Section: Previous Workmentioning

confidence: 99%

Supervised Learning of Quantizer Codebooks by Information Loss Minimization

Lazebnik¹,

Raginsky²

2009

IEEE Trans. Pattern Anal. Mach. Intell.

195

151

View full text Add to dashboard Cite

Abstract-This paper proposes a technique for jointly quantizing continuous features and the posterior distributions of their class labels based on minimizing empirical information loss, such that the quantizer index of a given feature vector approximates a sufficient statistic for its class label. Informally, the quantized representation retains as much information as possible for classifying the feature vector correctly. We derive an alternating minimization procedure for simultaneously learning codebooks in the Euclidean feature space and in the simplex of posterior class distributions. The resulting quantizer can be used to encode unlabeled points outside the training set and to predict their posterior class distributions, and has an elegant interpretation in terms of lossless source coding. The proposed method is validated on synthetic and real datasets, and is applied to two diverse problems: learning discriminative visual vocabularies for bag-of-features image classification, and image segmentation.

show abstract

“…In particular, we are interested in fitting Gaussian mixture models to data within a VQ framework [1]. In this approach, the PDF of an input vector X is represented as a weighted collection of Gaussians:…”

Section: Pixel Color Clustering and Kl Divergencementioning

confidence: 99%

Automatic Skin Pixel Selection and Skin Color Classification

Yoon¹,

Harville

Baker

et al. 2006

2006 International Conference on Image Processing

View full text Add to dashboard Cite

We describe an automatic, low-cost method for classifying skin color, independent of lighting and imaging device characteristics, using consumer digital cameras and a simple color calibration target. After color normalization and face detection are performed as described in prior work, pixels within the face region are clustered in color space in an unsupervised fashion, and a Gaussian mixture model (GMM) of the person's skin color is formed from the pixels belonging to the densest clusters containing at least some minimum fraction of the total pixels. This technique allows accurate modeling of non-uniformities in skin tone that are common in individuals, while avoiding contamination from shadows, specularities, eyes, lips, hair, and background. We incorporate these models into a skin color classification framework with improved performance over prior work. Given a set of exemplar face images with skin color labels assigned by an expert, we predict the label that would be chosen by the same expert for a new, test image by comparison of the respective GMMs of the test image and each exemplar. Specifically, we select the label of the exemplar image whose GMM has smallest Kullback-Leibler divergence from that of the test image.

show abstract

Lloyd clustering of Gauss mixture models for image compression and classification

Cited by 44 publications

References 55 publications

Hyper-Parameter in Hidden Markov Random Field

Hyper-Parameter in Hidden Markov Random Field

Supervised Learning of Quantizer Codebooks by Information Loss Minimization

Automatic Skin Pixel Selection and Skin Color Classification

Contact Info

Product

Resources

About