Selective scene modeling

This paper presents a new technique for region-oriented video compression based on the minimization of a minimum description length (MDL) criterion. The proposed scheme is based on a region-based motion compensation approach, in which the displaced frame difference (DFD) is coded together with the segmentation map and the motion parameters. The region-based segmentation is approximated by a polygonal representation which permits us to significantly reduce the segmentation coding cost. The polygonal approximation, which minimizes the MDL criterion, is considered the optimal approximation in terms of compression performance. In this contest, the MDL criterion has been developed and tested for the three following DFD coding techniques: (1) entropy coding of the quantized DFD image; (2) entropy coding of the quantized DFD values obtained after morphological filtering of the DFD image; and (3) DCT-based coding of the DFD image. The experimental results, presented on real video sequences, validate the proposed approach.

Proper scale for modeling visual data

We present a novel approach for clip-based key frame extraction. Our framework allows both clips with subtle changes as well as clips containing rapid shot changes, fades and dissolves to be well approximated. We show that creating key frame video abstractions can be achieved by transforming each frame of a video sequence into an eigenspace and then clustering this space using Gaussian mixture models (GMMs). A minimum description length (MDL) criterion is then used to determine the optimal number of GMM components to use in the clustering. The image nearest to the centres of each of the GMM components are selected as key frames. Unlike previous work, this technique relies on global video clip properties and results show that the key frames extracted give a very good representation of the overall clip content. We demonstrate the application of this technique on a database of 307 clips of wildlife footage containing dissolves, shot changes, fades, pans, zooms and a wide range of animal behaviours.

In regards to difficult selection of a threshold in Wavelet speech enhancement algorithm, a new adaptive threshold algorithm based on minimum description length criterion is proposed in this paper. The algorithm is a completely data-driven method and has very strong adaptability. It has characteristics of no requirement for prior knowledge of noise level and nature, preset threshold and choosing threshold behavior. In comparision with the wavelet soft and hard threshold algorithms, the proposed algorithm offers more enhanced speech with less distortion, and the simulation results have demonstrated the improvement in output SNR of enhanced speech, retained energy ratio, mean- squared error, and compression error ratio.

This paper presents a new technique for region-oriented video compression based on the minimization of a minimum description length (MDL) criterion. The proposed video compression scheme is based on a region-based motion compensation approach in which the displaced frame difference is partially coded together with the segmentation and the motion parameters. The region-based segmentation is approximated by a polygonal representation which permits to significantly reduce the segmentation coding cost. The polygonal approximation which minimizes the MDL criterion is finally considered as optimal. Experimental results are presented on real video sequences.

While non-negative blind source separation (nBSS) has found many successful applications in science and engineering, model order selection, determining the number of sources, remains a critical yet unresolved problem. Various model order selection methods have been proposed and applied to real-world data sets but with limited success, with both order over- and under-estimation reported. By studying existing schemes, we have found that the unsatisfactory results are mainly due to invalid assumptions, model oversimplification, subjective thresholding, and/or to assumptions made solely for mathematical convenience. Building on our earlier work that reformulated model order selection for nBSS with more realistic assumptions and models, we report a newly and formally revised model order selection criterion rooted in the minimum description length (MDL) principle. Adopting widely invoked assumptions for achieving a unique nBSS solution, we consider the mixing matrix as consisting of deterministic unknowns, with the source signals following a multivariate Dirichlet distribution. We derive a computationally efficient, stochastic algorithm to obtain approximate maximum-likelihood estimates of model parameters and apply Monte Carlo integration to determine the description length. Our modeling and estimation strategy exploits the characteristic geometry of the data simplex in nBSS. We validate our nBSS-MDL criterion through extensive simulation studies and on four real-world data sets, demonstrating its strong performance and general applicability to nBSS. The proposed nBSS-MDL criterion consistently detects the true number of sources, in all of our case studies.

Much research has focused on the problem of estimating the model order of autoregressive moving average (ARMA) processes. The most well-known of the proposed solutions for this problem include the final prediction error (FPE), Akaike information criterion (AIC), and minimum description length (MDL). A new approach for model order determination based on the MDL criterion is proposed and shown to depend on the minimum eigenvalue of a covariance matrix derived from the observed data. As a result, a new selection procedure for estimating the model order via MDL is proposed. Examples that illustrate the significantly improved accuracy of the proposed technique are given.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

An approach for using minimum description length (MDL) criterion for autoregressive moving-average (ARMA) models and autoregressive models with exogenous input (ARX) is formulated in terms of the minimum eigenvalue of a covariance matrix. It is shown that as the number of observation data points increases without bound, the MDL criterion reduces to a simple monotonic nonlinear transformation of this minimum eigenvalue. Consequently, it is shown that in the asymptotic case the minimum-eigenvalue provides exactly the same information as the original MDL formulation. Additionally, the approach does not require prior estimation of the model parameters, and thus has the potential for significantly reduced computation compared to other MDL-based model order determination schemes. The approach's application to the selection of ARMA and ARX model orders is discussed, and several numerical examples are presented. >

We investigate the problem of designing adaptive sequential linear predictors for the class of piecewise autoregressive multidimensional signals, and adopt an approach of minimum description length (MDL) to determine the order of the predictor and the support on which the predictor operates. The design objective is to strike a balance between the bias and variance of the prediction errors in the MDL criterion. The predictor design problem is particularly interesting and challenging for multidimensional signals (e.g., images and videos) because of the increased degree of freedom in choosing the predictor support. Our main result is a new technique of sequentializing a multidimensional signal into a sequence of nested contexts of increasing order to facilitate the MDL search for the order and the support shape of the predictor, and the sequentialization is made adaptive on a sample by sample basis. The proposed MDL-based adaptive predictor is applied to lossless image coding, and its performance is empirically established to be the best among all the results that have been published till present.

Blind source separation algorithm is usually not able to estimate the number of unknown signal sources. In many occasions, the number of source signal is unknown and may even be in dynamic changes. This paper has achieved to estimate the number of sources and real-time tracking using subspace method in the over-determined blind source separation, while the number of sources is unknown and dynamic . The first section is the estimation of the rank of signal subspace and the second section is about the subspace tracking algorithm. The subspace method is to separate the observed sensor signals into signal subspace and noise subspace. This will not only greatly reduce the noise, but also can estimate the number of active source signals by the measurement of eigenvalues. To achieve the real-time adjustment of the threshold in the dynamic blind source processing with Akaike’s information criterion (AIC) and the minimum description length criterion (MDL).

In this paper, an adaptive subspace classification technique using a fuzzy support vector machine (FSVM) is proposed for subspace-based TOA estimation in a multipath channel. For the subspace-based algorithm to work properly, the signal subspace should be correctly estimated from a data matrix composed of the received samples. Through SVD on the data matrix, singular values and vectors are derived in pairs. According to the derived singular values, the corresponding singular vectors are selected as a set of basis-spanning signal subspaces. Criteria such as the Akaike information criterion (AIC) and minimum description length criterion (MDLC) have previously been adopted for subspace classification. Instead of using these criteria, we propose an adaptive classification technique using a FSVM for improved subspace-based TOA estimation in a multipath channel.

The problem of modeling polynomial-phase signals is considered. Techniques based on the bootstrap and multiple hypotheses tests for optimal model selection of constant amplitude polynomial-phase signals embedded in stationary noise are developed. Phase parameter estimators based on both the least-squares method and the polynomial phase transform are used. The proposed techniques are compared with existing ones, including Akaike's (1978, 1974) information criterion and Rissanen's (1978) minimum description length criterion and are shown to outperform these procedures for the considered small sample sizes.

An important practical question, given various sources of noise in resting-state blood-oxygen-level-dependent (BOLD) functional magnetic resonance imaging (fMRI) images, is how much data to acquire to get identifiable and stable independent component analysis (ICA) networks in MRI scans. In this study, we investigate how ICA dimensionality, stability of the estimated independent components, and the spatiotemporal features of the extracted networks are affected by scan length. Our results show that the length of scan clearly influences the network dimensionality, peak voxel location of the spatial maps, spectral power, and functional network connectivity (FNC). Using minimum description length criteria and entropy-rate based order selection, we observed that the dimensionality decreased as the scan length decreased from 12 to 2 minutes. In addition, due to decreased dimensionality in shorter scans, some networks such as the ventral attention and subcortical networks could not be extracted. But the networks that could be identified in shorter scans including the default-mode, salience, and self-referential networks were stable (cluster quality index Iq > 0.8). Kurtosis of time-courses and spatial maps, as well as the spectral parameters (the dynamic range and fractional of amplitude of low frequency fluctuation; fALFF), and FNC also changed with the scan length with shorter scans having lower values. Consequently, scan length choice can have a significant impact on the of ICA-based restingstate fMRI studies. Future studies using larger sample sizes are warranted to further assess the implications of the scan length and data size on ICA outcome measures and reliability.

The combination of nonlocal grouping and transformed domain filtering has led to the state-of-the-art denoising techniques. In this paper, we extend this line of study to the denoising of synthetic aperture radar (SAR) images based on clustering the noisy image into disjoint local regions with similar spatial structure and denoising each region by the linear minimum mean-square error (LMMSE) filtering in principal component analysis (PCA) domain. Both clustering and denoising are performed on image patches. For clustering, to reduce dimensionality and resist the influence of noise, several leading principal components identified by the minimum description length criterion are used to feed the K-means clustering algorithm. For denoising, to avoid the limitations of the homomorphic approach, we build our denoising scheme on additive signal-dependent noise model and derive a PCA-based LMMSE denoising model for multiplicative noise. Denoised patches of all clusters are finally used to reconstruct the noise-free image. The experiments demonstrate that the proposed algorithm achieved better performance than the referenced state-of-the-art methods in terms of both noise reduction and image detail preservation.

The authors propose a two-dimensional direction-of-arrival (DOA) estimation for multi-path environments, in which there are uncorrelated, partially correlated and coherent signals. The inability to identify non-coherent signals from coherent ones results in a considerable waste of sensors. In this work, an adaptive and automated threshold is considered for the efficient separation of noise, non-coherent signals, and coherent groups. First, non-coherent signals, coherent groups, and noise subspace are separated using k-medoids clustering. After determining the number of sources, non-coherent signals and coherent groups, non-coherent DOAs are estimated separately. The number of coherent signals in each coherence group is determined by the minimum descriptive length criterion. Finally, coherent DOAs are estimated in each group by constructing a coherent estimation matrix. The proposed method does not require any prior information such as knowing the number of signals or the covariance matrix of uncorrelated signals. The simulation results show that the proposed method is able to distinguish between the non-coherent and coherent signals, even at low signal-to-noise ratios and a small number of snapshots. Also, in terms of detection probability and estimation accuracy, it shows an improvement of over 1.2 and 83%, respectively, compared with the conventional forward−backward spatial smoothing scheme.