Y. Raja, S. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In 5th European Conference on Computer Vision, pages 460--474, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....one Gaussian is equivalent to one state of a HMM. At first we use an Epanechnikov s kernel estimation [10] i.e to place an isotropic bump at each observation vector. Hence, we work only with the data and avoid the problem with empty bins. Then, We use the EM algorithm with splitting components [8] to obtain a better estimation of each pdf. The probabilistic model is built as follows: p(x #) g=1 # g p g (x # g ) 1) The # l are the mixing weights ( i=1 # i = 1) the mixture parameters are denoted #=# 1 , #N ,# 1 , # N and modeled as Gaussians of d dimensional vectors. ....

....of each pdf. The probabilistic model is built as follows: p(x #) g=1 # g p g (x # g ) 1) The # l are the mixing weights ( i=1 # i = 1) the mixture parameters are denoted #=# 1 , #N ,# 1 , # N and modeled as Gaussians of d dimensional vectors. The approach of Raja et al. [8] is to automatically determine the number of Gaussians by EM. The reponsability r m of each Gaussian blob m is defined as: r m = # l #m p(x l # m ) p(x l #) k=argmin m (r m ) 2) The blob with the lowest responsibility is split in two as this blob does not seem meaningful enough. The ....

Y. Raja, S. J. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In ECCV98, 1998.

.... change in the presence of specularities [8] Colour transformations like HSV and HLS that convert image RGB values to a hue based representation [17] allow not only for a more intuitive description of colour but can also be used in applications such as object recognition [19, 8] and face tracking [16]. Colour naming, the division of colour space into regions identified by colour names, is closely linked with the concept of hue, and has been successfully used in image retrieval [13] and visual surveillance [2] Remarkably, the human vision system can ascribe fairly constant hues to surfaces ....

Y. Raja, S. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In 5th European Conference on Computer Vision, pages 460--474, 1998.

....that clipping does not occur. 21 colour lies close to the Planckian locus (it is desaturated pink) However, illuminant estimation from faces is a particularly valuable task because skin colour changes dramatically with changing illumination and so complicates face recognition [21] and tracking [18]. In Figure 12 a face viewed under 4200K is shown together with a plot showing the dichromatic intersection (based on the combined RGB pixels of hand segmented regoins from the forehead and the cheeks) that yields the estimated illuminant. Again, our algorithm manages to provide an estimate close ....

Y. Raja, S. McKenna, and S. Gong. Colour model selection and adapta- tion in dynamic scenes. In Proceedings European Conference Computer Vision, Freiburg, Germany, 1998.

.... for our algorithm since skin colour lies close to the PlanckJan locus (it is desaturated pink) However, illuminant estimation from faces is a particularly valuable task because skin colour changes dramatically with changing illumination and so complicates face recognition [12] and tracking [10]. In Figure 9 a face viewed under 4200K is shown together with a plot showing the dichromatic intersection that yields the estimated illuminant. Again, our algorithm manages to provide an estimate close to the actual illuminant, the error in terms of correlated colour temperature difference is ....

Y. Raja, S. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In Proceedings European Conference Computer Vision, l'eiburg, Germany, 1998.

....to track fast and reliably a moving face [2] Variuous improvements to this constancy model have been proposed in order to deal with small changes in the colour of the illumination. These consist on tracking the motion of the skin colour cluster in rg space by using stochastic prediction models [9] or by clustering [13] These algoritms have a problem in common: they can not deal with sudden changes in lighting colour. Other algorithms that also work in rg normalised space deal with sudden changes in lighting by matching the colour distributions by a shift in illuminant colour [1] ....

Y. Raja, S.J. McKenna, S. Gong. Colour model selection and adaptation in dynamic scenes. Proc. ECCV. Vol. I, 460-474. 1998.

.... m s ; s ; I r g b ) can not be rejected. On the other hand, it is not possible to nd an analytic model for the background pdf, so we will model it with a uniform distribution, h b (I r g b ) Other authors have indicated di erent preferences for modelling the colour distributions. In [11] Gaussian mixture models, whereas in [5] and [12] pure histogram based representations are chosen. In our experiments we found that using a a continuous model yields better results because of the high space dimensionality (3D) 1.5 2.5 0.5 1 1.5 2 2.5 0.5 1.5 2.5 R GW G GW B GW 0 5 ....

Y. Raja, S.J. McKenna, S. Gong. Colour model selection and adaptation in dynamic scenes. Proc. ECCV. Vol. I, 460-474. 1998.

....a blob restricts the blob to be of a single color which is not a general enough assumption about the clothes people wear which are patterned and have many colors. Fitting a mixture of Gaussian using the EM algorithm provides a way to model blobs with a mixture of colors. This technique was used in [11, 12] for color based tracking of a single blob and was applied to tracking faces. Mixture of Gaussian techniques face the problem of choosing the right number of Gaussians for the assumed model. Non parametric techniques using histograms have also been used in [13] In this work they used ....

Y. Raja, S. J. Mckenna, and S. Gong, "Colour model selection and adaptation in dynamic scenes," in 5th European Conference of Computer Vision, 1998.

....a feature that is robust to partial occlusion, scaling and object deformation. It is also relatively stable under rotation in depth in certain applications. Therefore color distributions have been used successfully to track nonrigid bodies [17, 2, 14, 6] with applications like tracking heads [1, 6, 13, 14], hands [11] and other body parts against cluttered backgrounds from stationary or moving platforms. Color distributions have also been used for object recognition. A variety of parametric and non parametric statistical techniques have been used to model the color distribution of a homogeneous ....

....to model regions with mixtures of colors. For example, people s clothing and surfaces with texture usually contain patterns and mixture of colors. Fitting a mixture of Gaussians using the EM algorithm provides a way to model color blobs with a mixture of colors. This technique was used in [13, 14] for color based tracking of a single blob and was applied to tracking faces. The mixture of Gaussians technique faces the problem of choosing the right number of Gaussians for the assumed model (model selection) Nonparametric techniques using histograms have been widely used for modeling ....

Y. Raja, S. J. Mckenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In Proc. 5th European Conference of Computer Vision, 1998.

....The aspiration of having good pattern recognition feature prompts the work described in this paper. Several previous work on adapting feature models in tracking include: Mckenna [1] uses adaptive mixture of Gaussian models and log likelihood measurements to perform color tracking. Raja [2] uses color mixture models and adaptation in real time tracking as well. The updates in Gaussian mixture models, albeit sometime inadequate, involve very few parameters and thus are amenable to fast processing. Ying [3] uses adaptive SOM (self organizing map) techniques to model object color ....

Y. Raja, S.J. McKenna, and S. Gong, "Colour model selection and adaptation in dynamic scenes," in ECCV, 1998.

....of a blob were modeled using a single Gaussian in the three dimension Y UV space. The spatial properties of a blob were modeled using pixel support maps. Fitting a mixture of Gaussian using the EM algorithm provides a way to model blobs with a mixture of colors. This technique was used in [7, 8] for color based tracking of a single blob and was applied to tracking faces. Mixture of Gaussian techniques face the problem of choosing the right number of Gaussian for the model. Non parametric techniques using histograms have also been used in [6] In this work they used 3 dimensional adaptive ....

Y. Raja, S. J. Mckenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In 5th European Conference of Computer Vision, 1998.

....a feature that is robust to partial occlusion, scaling and object deformation. It is also relatively stable under rotation in depth in certain applications. Therefore color distributions have been used successfully to track nonrigid bodies [17, 2, 14, 6] with applications like tracking heads [1, 6, 13, 14], hands [11] and other body parts against cluttered backgrounds from stationary or moving platforms. Color distributions have also been used for object recognition. A variety of parametric and non parametric statistical techniques have been used to model the color distribution of a homogeneous ....

....to model regions with mixtures of colors. For example, people s clothing and surfaces with texture usually contain patterns and mixture of colors. Fitting a mixture of Gaussians using the EM algorithm provides a way to model color blobs with a mixture of colors. This technique was used in [13, 14] for color based tracking of a single blob and was applied to tracking faces. The mixture of Gaussians technique faces the problem of choosing the right number of Gaussians for the assumed model (model selection) Nonparametric techniques using histograms have been widely used for modeling the ....

Y. Raja, S. J. Mckenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In Proc. 5th European Conference of Computer Vision, 1998.

....of a blob were modeled using a single Gaussian in the three dimension Y UV space. The spatial properties of a blob were modeled using pixel support maps. Fitting a mixture of Gaussian using the EM algorithm provides a way to model blobs with a mixture of colors. This technique was used in [7, 8] for color based tracking of a single blob and was applied to tracking faces. Mixture of Gaussian techniques face the problem of choosing the right number of Gaussian for the model. Non parametric techniques using histograms have also been used in [6] In this work they used 3 dimensional adaptive ....

Y. Raja, S. J. Mckenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In 5th European Conference of Computer Vision, 1998.

....one threshold can be easily found in two peak histogram that corresponds to simple background, it is still hard to handle cluttered background because finding good thresholds can be very complicated. Another approach is to make parametric color models by Gaussian distribution or Gaussian Mixtures [8, 4]. The problem is that there is not enough prior knowledge to determine the number of components of the distribution in advance. Our color segmentation scheme is to approximate the color distribution of an image in the HSI color space by 1 D self organizing map (SOM) in which each output neuron ....

....a simple color classifier by competition among its output neurons, through which the image at current time frame can be segmented. However, this classifier may not be good for the next time frame because of the non stationary density of color. Model adaptation over time was ever addressed in [8], in which a Gaussian mixture model was used, and a linear extrapolation was employed to adjust the parameters of the model by a set of labeled training data drawn from the new frame. However, since the new image is not segmented, these labeled data set is hard to obtain. Our solution to this ....

Y.Raja, S.McKenna, S.Gong, "Colour Model Selection and Adaptation in Dynamic Scenes", Proc. ECCV'98, 1998

....out. According to the representation of color distribution in certain color spaces, current techniques of color tracking can be classified into two general approaches: non parametric (Swain and Ballard 1991; Kjeldsen and Kender 1996; Jones and Rehg 1998; Wu, Liu, and Huang 2000) and parametric (Raja, McKenna, and Gong 1998). Many different color spaces, such as RGB, HSV, N RGB, have been used in current research. However, Copyright c # 2000, American Association for Artificial Intelligence (www.aaai.org) All rights reserved. many of these techniques are plagued by some special difficulties such as large ....

Raja, Y.; McKenna, S.; and Gong, S. 1998. Colour model selection and adaptation in dynamic scenes. In Proc. of European Conf. on Computer Vision.

....localize and track hand and face. The core of color tracking is color based segmentation. According to the representation of color distribution in certain color spaces, current techniques of color tracking can be classified into two general approaches: non parametric [12, 8, 7, 15] and parametric [14, 9, 16]. One of the non parametric approaches is based on color histograms [12, 8, 7] Since color space is quantized by the structure of the histogram, this technique shares the same problem with non parametric density estimation, in which the level of quantization will affect the estimation. How to ....

....growing, pruning and merging schemes. Generally, these non parametric approaches work e#ectively when the quantization level is properly set and there are su#cient data. Parametric approaches model the color density in parametric forms such as Gaussian distribution or Gaussian Mixture models [14, 9, 16]. ExpectationMaximization (EM) o#ers a way to fit probabilistic models to the observation data. The di#culty of model order selection could be handled by heuristics [9] or cross validation. However, when we try to apply these techniques to track human hand and face in some virtual environment ....

[Article contains additional citation context not shown here]

Y.Raja, S.McKenna and S.Gong, "Colour Model Selection and Adaptation in Dynamic Scenes", Proc. ECCV'98, 1998

No context found.

Y. Raja, S. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. ECCV, Freiburg, Germany, 1998.

No context found.

Y. Raja, S. McKenna, S. Gong, Colour model selection and adaptation in dynamic scenes, European Conference on Computer Vision, Freiburg, Germany, 1998.

....[5] provide effective tools for the determination of mixture components. However, the resulting mixture models depend on the a priori knowledge of the number of mixtures. The model order can be determined using constructive algorithms that employ cross validation techniques for model training [10]. However the disadvantage of such methods is that they require a validation set, which is often not available. Alternative approaches to determine the number of clusters are based on information criteria, such as A Information Criterion (AIC) 1] Bayesian Information Criterion (BIC) 14] and ....

....component identifier, labels at the bottom left indicate the gesture model identifier whereas labels at the top left indicate the number of extracted sequences corresponding to each gesture model. 5. 1 Determining the Number of Gesture Models Looking at components [6,23] 7,8,12] 9,18] [10,16] in Figure 9 we can see a general tendency of MDL to overestimate the total number of atomic components. This results in identical gestures being described by multiple models containing different, however similar atomic components as can be seen in Figure 10 for models [2,7,11,12,13] 3,9] ....

[Article contains additional citation context not shown here]

Y. Raja, S. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. ECCV, Freiburg, Germany, 1998.

No context found.

Y. Raja, S. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In 5th European Conference on Computer Vision, pages 460--474, 1998.

No context found.

Yogesh Raja, Stephen J. Mckenna, and Shaogang Gong, "Colour model selection and adaptation in dynamic scenes," in Proc. 5th European Conference of Computer Vision, 1998.

No context found.

Y. Raja, S. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In ECCV, 1998.

No context found.

Y. Raja, S. McKenna, and S. Gong, "Colour model selection and adaptation in dynamic scenes," in Proc. of European Conf. on Computer Vision, 1998, pp. 460--475.

No context found.

Y. Raja, S. McKenna, and S. Gong, "Colour Model Selection and Adaptation in Dynamic Scenes," in Proc. ECCV, 1998.

No context found.

Y. Raja, S. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In Proceedings of the European Conference on Computer Vision, 1998.

No context found.

Y. Raja, S. J. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In ECCV, pages 460--474, 1998.

No context found.

Raja Y, McKenna SJ, Gong S (1998) Colour model selection and adaptation in dynamic scenes. Proc. European Conf. on Computer Vision, pp 460-474

No context found.

Y. Raja, S. J. McKenna, and S. Gong. Colour model selection and adaptation in dynamic scenes. In ECCV, pages 460--474, 1998.

No context found.

Yogesh Raja, J. McKenna, and Shaogang Gong. Colour model selection and adaptation in dynamic scenes. In The Fifth European Conference on Computer Vision. European Vision Society, 1998.

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