This paper proposes a global stereo correspondence using robust matching likelihoods and minimum spanning tree (MST) leveraged smooth priors in a probabilistic graphical model framework. The matching likelihoods of ...
详细信息
ISBN:
(纸本)9781467321969
This paper proposes a global stereo correspondence using robust matching likelihoods and minimum spanning tree (MST) leveraged smooth priors in a probabilistic graphical model framework. The matching likelihoods of the stereo correspondence can be robustly constructed as data term by aggregating initial matching costs from Weber local descriptors using an unsymmetrical guided filtering in a linear model. The disparity priors are devised as smooth term to characterize the smoothness constraints leveraged by the MST structure. The presented stereo approach provides an effective and efficient way to reflect robust visual dissimilarity and resolve local and regional discontinuities. Experiments demonstrate that the proposed global stereo matching method can produce piecewise smooth, accurate and dense disparity map, while removing effectively the visual ambiguity of the stereo matching problem.
Deep Neural Network Hidden Markov Models, or DNN-HMMs, are recently very promising acoustic models achieving good speech recognition results over Gaussian mixture model based HMMs (GMM-HMMs). In this paper, for emotio...
详细信息
Deep Neural Network Hidden Markov Models, or DNN-HMMs, are recently very promising acoustic models achieving good speech recognition results over Gaussian mixture model based HMMs (GMM-HMMs). In this paper, for emotion recognition from speech, we investigate DNN-HMMs with restricted Boltzmann Machine (RBM) based unsupervised pre-training, and DNN-HMMs with discriminative pre-training. Emotion recognition experiments are carried out on these two models on the eNTERFACE'05 database and Berlin database, respectively, and results are compared with those from the GMM-HMMs, the shallow-NN-HMMs with two layers, as well as the Multi-layer Perceptrons HMMs (MLP-HMMs). Experimental results show that when the numbers of the hidden layers as well hidden units are properly set, the DNN could extend the labeling ability of GMM-HMM. Among all the models, the DNN-HMMs with discriminative pre-training obtain the best results. For example, for the eNTERFACE'05 database, the recognition accuracy improves 12.22% from the DNN-HMMs with unsupervised pre-training, 11.67% from the GMM-HMMs, 10.56% from the MLP-HMMs, and even 17.22% from the shallow-NN-HMMs, respectively.
This paper proposes a novel global stereo matching method using aggregated likelihoods and multi-scale priors. The likelihoods of dense stereo correspondences as data term can be robustly expressed by aggregated match...
详细信息
We present a new open source toolkit for phrase-based and syntax-based machine translation. The toolkit supports several state-of-the-art models developed in statistical machine translation, including the phrase-based...
详细信息
A new prediction algorithm of tourists flow distribution based on transition probability matrix (TPM) is proposed in this paper. In order to analyze the visitor transition-behavior and the tourists distribution model,...
详细信息
In this paper, a robust homography estimation method is proposed to match multiview images in the uncalibrated case. This method formulates a new loss function to verify homography hypothesis, which combines models of...
详细信息
The H approach is used for the clearance of flight control laws when some parameters in the flight control system vary in a certain range. The proposed H approach is developed from general H theory and is applied to t...
详细信息
Distributed video coding (DVC) is a novel video coding paradigm. One approach to DVC is Wyner-Ziv distributed video coding. The accuracy of the correlation noise model can influence the performance of the video coder ...
详细信息
Distributed video coding (DVC) is a novel video coding paradigm. One approach to DVC is Wyner-Ziv distributed video coding. The accuracy of the correlation noise model can influence the performance of the video coder directly. In order to enhance the accuracy of the distribution model, EM algorithm based mixture Laplace-uniform distribution model and basic Laplace-uniform distribution model for DCT alternating current coefficients are established. Then the model is selected adaptively using fuzzy inference. Experimental results suggest that the proposed mixture correlation noise model can describe the heavy tail and sudden change of the noise accurately at high rate and make significant improvement on the coding efficiency compared with the DISCOVER's noise model. Meanwhile, fuzzy inference based adaptive noise model selection method can reduce the operation complexity to some extent, while not influencing rate-distortion performance.
Conventional single-chip digital cameras use color filter arrays(CFA) to sample different spectral components. image demosaicing is a problem of interpolating these data to complete red, green, and blue values for eac...
Conventional single-chip digital cameras use color filter arrays(CFA) to sample different spectral components. image demosaicing is a problem of interpolating these data to complete red, green, and blue values for each image pixel, to produce an RGB image. Many color demosaicing(CDM) methods assume that the high local spatial redundancy exists among the color samples. Such an assumption, however, may be fail for images with high color saturation and sharp color transitions. This paper presents an adaptive demosaicing algorithm by exploiting both the non-local similarity and the local correlation(NLS-LC) in the color filter array image. First, the most flattest nonlocal image patches are searched in the searching window centered on the estimated pixel. Second, the patch, which is the most similar to the current patch, is selected among the most smoothest nonlocal patches. Third, according to the similar degree and the local correlation degree, the obtained nonlocal image patch and the current patch are adaptively chosen to estimate the missing color samples. Experimental results indicate that the proposed method exhibits superior performance over many state-of-the-art color interpolation methods.
暂无评论