arXiv subjects

Brani Vidakovic

Publications and source records attributed to Brani Vidakovic.

31 records · Page 2Linked to original sources

arXiv2022

Gamma-Minimax Wavelet Shrinkage with Three-Point Priors

In this paper we propose a method for wavelet denoising of signals contaminated with Gaussian noise when prior information about the $L^2$-energy of the signal is available. Assuming the independence model, according to which the wavelet coefficients are treated individually, we propose a simple, level dependent shrinkage rules that turn out to be $Γ$-minimax for a suitable class of priors. The proposed methodology is particularly well suited in denoising tasks when the signal-to-noise ratio is low, which is illustrated by simulations on the battery of standard test functions. Comparison to some standardly used wavelet shrinkage methods is provided.

stat.ME↗

arXiv2022

Non-decimated 2D Wavelet Spectrum and Its Use in Breast Cancer Diagnostics

To improve diagnostic accuracy of breast cancer detection, several researchers have used the wavelet-based tools, which provide additional insight and information for aiding diagnostic decisions. The accuracy of such diagnoses, however, can be improved. This paper introduces a wavelet-based technique, non-decimated wavelet transform (NDWT)-based scaling estimation, that improves scaling parameter estimation over the traditional methods. One distinctive feature of NDWT is that it does not decimate wavelet coefficients at multiscale levels resulting in redundant outputs which are used to lower the variance of scaling estimators. Another interesting feature of the proposed methodology is the freedom of dyadic constraints for inputs, typical for standard wavelet-based approaches. To compare the estimation performance of the NDWT method to a conventional orthogonal wavelet transform-based method, we use simulation to estimate the Hurst exponent in two-dimensional fractional Brownian fields. The results of the simulation show that the proposed method improves the conventional estimators of scaling and yields estimators with smaller mean-squared errors. We apply the NDWT method to classification of mammograms as cancer or control and, for publicly available mammogram images from the database at the University of South Florida, find the the diagnostic accuracy in excess of 80%.

stat.ME↗

arXiv2022

Robust Wavelet-based Assessment of Scaling with Applications

A number of approaches have dealt with statistical assessment of self-similarity, and many of those are based on multiscale concepts. Most rely on certain distributional assumptions which are usually violated by real data traces, often characterized by large temporal or spatial mean level shifts, missing values or extreme observations. A novel, robust approach based on Theil-type weighted regression is proposed for estimating self-similarity in two-dimensional data (images). The method is compared to two traditional estimation techniques that use wavelet decompositions; ordinary least squares (OLS) and Abry-Veitch bias correcting estimator (AV). As an application, the suitability of the self-similarity estimate resulting from the the robust approach is illustrated as a predictive feature in the classification of digitized mammogram images as cancerous or non-cancerous. The diagnostic employed here is based on the properties of image backgrounds, which is typically an unused modality in breast cancer screening. Classification results show nearly 68% accuracy, varying slightly with the choice of wavelet basis, and the range of multiresolution levels used.

stat.ME↗

arXiv2019

Non-decimated Quaternion Wavelet Spectral Tools with Applications

Quaternion wavelets are redundant wavelet transforms generalizing complex-valued non-decimated wavelet transforms. In this paper we propose a matrix-formulation for non-decimated quaternion wavelet transforms and define spectral tools for use in machine learning tasks. Since quaternionic algebra is an extension of complex algebra, quaternion wavelets bring redundancy in the components that proves beneficial in wavelet based tasks. Specifically, the wavelet coefficients in the decomposition are quaternion-valued numbers that define the modulus and three phases. The novelty of this paper is definition of non-decimated quaternion wavelet spectra based on the modulus and phase-dependent statistics as low-dimensional summaries for 1-D signals or 2-D images. A structural redundancy in non-decimated wavelets and a componential redundancy in quaternion wavelets are linked to extract more informative features. In particular, we suggest an improved way of classifying signals and images based on their scaling indices in terms of spectral slopes and information contained in the three quaternionic phases. We show that performance of the proposed method significantly improves when compared to the standard versions of wavelets including the complex-valued wavelets. To illustrate performance of the proposed spectral tools we provide two examples of application on real-data problems: classification of sounds using scaling in high-frequency recordings over time and monitoring of steel rolling process using the fractality of captured digitized images. The proposed tools are compared with the counterparts based on standard wavelet transforms.

stat.AP↗

arXiv2019

Non-decimated Complex Wavelet Spectral Tools with Applications

In this paper we propose spectral tools based on non-decimated complex wavelet transforms implemented by their matrix formulation. This non-decimated complex wavelet spectra utilizes both real and imaginary parts of complex-valued wavelet coefficients via their modulus and phases. A structural redundancy in non-decimated wavelets and a componential redundancy in complex wavelets act in a synergy when extracting wavelet-based informative descriptors. In particular, we suggest an improved way of separating signals and images based on their scaling indices in terms of spectral slopes and information contained in the phase in order to improve performance of classification. We show that performance of the proposed method is significantly improved when compared with procedures based on standard versions of wavelet transforms or on real-valued wavelets. It is also worth mentioning that the matrix-based non-decimated wavelet transform can handle signals of an arbitrary size and in 2-D case, rectangular images of possibly different and non-dyadic dimensions. This is in contrast to the standard wavelet transforms where algorithms for handling objects of non-dyadic dimensions requires either data preprocessing or customized algorithm adjustments. To demonstrate the use of defined spectral methodology we provide two examples of application on real-data problems: classification of visual acuity using scaling in pupil diameter dynamic in time and diagnostic and classification of digital mammogram images using the fractality of digitized images of the background tissue. The proposed tools are contrasted with the traditional wavelet based counterparts.

stat.AP↗

arXiv2018

An Empirical approach to Survival Density Estimation for randomly-censored data using Wavelets

Density estimation is a classical problem in statistics and has received considerable attention when both the data has been fully observed and in the case of partially observed (censored) samples. In survival analysis or clinical trials, a typical problem encountered in the data collection stage is that the samples may be censored from the right. The variable of interest could be observed partially due to the presence of a set of events that occur at random and potentially censor the data. Consequently, developing a methodology that enables robust estimation of the lifetimes in such setting is of high interest for researchers. In this paper, we propose a non-parametric linear density estimator using empirical wavelet coefficients that are fully data driven. We derive an asymptotically unbiased estimator constructed from the complete sample based on an inductive bias correction procedure. Also, we provide upper bounds for the bias and analyze the large sample behavior of the expected $\mathbb{L}_{2}$ estimation error based on the approach used by Stute (1995), showing that the estimates are asymptotically normal and possess global mean square consistency. In addition, we evaluate the proposed approach via a theoretical simulation study using different exemplary baseline distributions with different sample sizes. In this study, we choose a censoring scheme that produces a censoring proportion of 40\% on average. Finally, we apply the proposed estimator to real data-sets previously published, showing that the proposed wavelet estimator provides a robust and useful tool for the non-parametric estimation of the survival time density function.

stat.AP↗

arXiv2018

Least Squares Wavelet-based Estimation for Additive Regression Models using Non Equally-Spaced Designs

Additive regression models are actively researched in the statistical field because of their usefulness in the analysis of responses determined by non-linear relationships with multivariate predictors. In this kind of statistical models, the response depends linearly on unknown functions of predictor variables and typically, the goal of the analysis is to make inference about these functions. In this paper, we study the problem of additive regression using a least squares approach based on periodic orthogonal wavelets on the interval [0,1]. For this estimator, we obtain strong consistency (with respect to the $\mathbb{L}_{2}$ norm) characterized by optimal convergence rates up to a logarithmic factor, independent of the dimensionality of the problem. This is achieved by truncating the model estimates by a properly chosen parameter, and selecting the multiresolution level $J$ used for the wavelet expansion, as a function of the sample size. In this approach, we obtain these results without the assumption of an equispaced design, a condition that is typically assumed in most wavelet-based procedures. Finally, we show practical results obtained from a simulation study and a real life application, demonstrating the applicability of the proposed methods for the problem of non-linear robust additive regression models.

stat.ME↗

arXiv2018

Empirical Wavelet-based Estimation for Non-linear Additive Regression Models

Additive regression models are actively researched in the statistical field because of their usefulness in the analysis of responses determined by non-linear relationships with multivariate predictors. In this kind of statistical models, the response depends linearly on unknown functions of predictor variables and typically, the goal of the analysis is to make inference about these functions. In this paper, we consider the problem of Additive Regression with random designs from a novel viewpoint: we propose an estimator based on an orthogonal projection onto a multiresolution space using empirical wavelet coefficients that are fully data driven. In this setting, we derive a mean-square consistent estimator based on periodic wavelets on the interval $[0,1]$. For construction of the estimator, we assume that the joint distribution of predictors is non-zero and bounded on its support; We also assume that the functions belong to a Sobolev space and integrate to zero over the $[0,1]$ interval, which guarantees model identifiability and convergence of the proposed method. Moreover, we provide the $\mathbb{L}_{2}$ risk analysis of the estimator and derive its convergence rate. Theoretically, we show that this approach achieves good convergence rates when the dimensionality of the problem is relatively low and the set of unknown functions is sufficiently smooth. In this approach, the results are obtained without the assumption of an equispaced design, a condition that is typically assumed in most wavelet-based procedures. Finally, we show practical results obtained from simulated data, demonstrating the potential applicability of our method in the problem of additive regression models with random designs.

stat.AP↗

arXiv2018

Bayesian nonparametric regression using complex wavelets

In this paper we propose a new adaptive wavelet denoising methodology using complex wavelets. The method is based on a fully Bayesian hierarchical model in the complex wavelet domain that uses a bivariate mixture prior on the wavelet coefficients. The heart of the procedure is computational, where the posterior mean is computed through Markov chain Monte Carlo (MCMC) simulations. We show that the method has good performance, as demonstrated by simulations on the well-known test functions and by comparison to a well-established complex wavelet-based denoising procedure. An application to real-life data set is also considered.

stat.ME↗

arXiv2017

Estimation of the Hurst Exponent Using Trimean Estimators on Nondecimated Wavelet Coefficients

Hurst exponent is an important feature summarizing the noisy high-frequency data when the inherent scaling pattern cannot be described by standard statistical models. In this paper, we study the robust estimation of Hurst exponent based on non-decimated wavelet transforms (NDWT). The robustness is achieved by applying a general trimean estimator on non-decimated wavelet coefficients of the transformed data. The general trimean estimator is derived as a weighted average of the distribution's median and quantiles, combining the median's emphasis on central values with the quantiles' attention to the extremes. The properties of the proposed Hurst exponent estimators are studied both theoretically and numerically. Compared with other standard wavelet-based methods (Veitch $\&$ Abry (VA) method, Soltani, Simard, $\&$ Boichu (SSB) method, median based estimators MEDL and MEDLA), our methods reduce the variance of the estimators and increase the prediction precision in most cases. The proposed methods are applied to a data set in high frequency pupillary response behavior (PRB) with the goal to classify individuals according to a degree of their visual impairment.

stat.ME↗

arXiv2017

MEDL and MEDLA: Methods for Assessment of Scaling by Medians of Log-Squared Nondecimated Wavelet Coefficients

High-frequency measurements and images acquired from various sources in the real world often possess a degree of self-similarity and inherent regular scaling. When data look like a noise, the scaling exponent may be the only informative feature that summarizes such data. Methods for the assessment of self-similarity by estimating Hurst exponent often involve analysis of rate of decay in a spectrum defined in various multiresolution domains. When this spectrum is calculated using discrete non-decimated wavelet transforms, due to increased autocorrelation in wavelet coefficients, the estimators of $H$ show increased bias compared to the estimators that use traditional orthogonal transforms. At the same time, non-decimated transforms have a number of advantages when employed for calculation of wavelet spectra and estimation of Hurst exponents: the variance of the estimator is smaller, input signals and images could be of arbitrary size, and due to the shift-invariance, the local scaling can be assessed as well. We propose two methods based on robust estimation and resampling that alleviate the effect of increased autocorrelation while maintaining all advantages of non-decimated wavelet transforms. The proposed methods extend the approaches in existing literature where the logarithmic transformation and pairing of wavelet coefficients are used for lowering the bias. In a simulation study we use fractional Brownian motions with a range of theoretical Hurst exponents. For such signals for which "true" $H$ is known, we demonstrate bias reduction and overall reduction of the mean-squared error by the two proposed estimators. For fractional Brownian motions, both proposed methods yield estimators of $H$ that are asymptotically normal and unbiased.

stat.ME↗

arXiv2016

A note on Bayesian wavelet-based estimation of scaling

A number of phenomena in various fields such as geology, atmospheric sciences, economics, to list a few, can be modeled as a fractional Brownian motion indexed by Hurst exponent $H$. This exponent is related to the degree of regularity and self-similarity present in the signal, and it often captures important characteristics useful in various applications. Given its importance, a number of methods have been developed for the estimation of the Hurst exponent. Typically, the proposed methods do not utilize prior information about scaling of a signal. Some signals are known to possess a theoretical value of the Hurst exponent, which motivates us to propose a Bayesian approach that incorporates this information via a suitable elicited prior distribution on $H$. This significantly improves the accuracy of the estimation, as we demonstrate by simulations. Moreover, the proposed method is robust to small misspecifications of the prior location. The proposed method is applied to a turbulence time series for which Hurst exponent is theoretically known by Kolmogorov's K41 theory.

stat.ME↗

arXiv2016

WavmatND: A MATLAB Package for Non-Decimated Wavelet Transform and its Applications

A non-decimated wavelet transform (NDWT) is a popular version of wavelet transforms because of its many advantages in applications. The inherent redundancy of this transform proved beneficial in tasks of signal denoising and scaling assessment. To facilitate the use of NDWT, we built a MATLAB package, {\bf WavmatND}, which has three novel features: First, for signals of moderate size the proposed method reduces computation time of the NDWT by replacing repetitive convolutions with matrix multiplications. Second, submatrices of an NDWT matrix can be rescaled, which enables a straightforward inverse transform. Finally, the method has no constraints on a size of the input signal in one or in two dimensions, so signals of non-dyadic length and rectangular two-dimensional signals with non-dyadic sides can be readily transformed. We provide illustrative examples and a tutorial to assist users in application of this stand-alone package.

stat.AP↗