Font Size: a A A

Research On Key Techniques Of Image Statistical Modeling And Noise Analysis

Posted on:2016-08-28Degree:DoctorType:Dissertation
Country:ChinaCandidate:C W TangFull Text:PDF
GTID:1108330503993775Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
This thesis deals with the problem of statistics based image analysis technique and its applications in various tasks of digital image processing, including image modeling and application in transform domain, image quality assessment and noise analysis and parameter estimation of natural image. Knowledge about statistics of natural images is crucial in many applications such as image compression, image denoising and target recognition. They are also important in understanding the biological vision system. Statistical methods can extract intrinsic properties from highly complex images, and incorporating of these properties is expected to improve the efficiency and performance of the image processing system, and hence we need to focus on researches of this area. Wavelet coefficients of image have outstanding features such as low entropy, multi-resolution and de-relevance, which are widely used in image compression, denoising, segmentation, texture analysis and synthesis. Therefore, a good description of wavelet coefficients statistics and its analytical representation can further reveal the image properties, thus providing prior knowledge that is indispensable for various image manipulations and improving their performances. As one of the most important applications of human visual system modeling, image quality assessment algorithms aim to predict subjective visual quality of images, so that can be used in the optimization of image processing systems which target to deliver images with better perceptual quality. As the most feasible aspect of image quality assessment, no reference method is characterized by its complexity. while natural image statistics is considered to be the most promising method to tackle with no reference image quality assessment problem. Noise analysis and parameter estimation of natural image have long been a widely studies topic due to its practical usage. Research on noise statistics and accurate estimation of its parameters(e.g. variance) not only directly affect the performance of image denoising algorithms, but also can be applied to the field of imaging system evaluation, camera calibration, trace detection, etc. Contents of this thesis can be divided into three Sections that are detailed as follows.Section 1 analyzes natural image modeling in transform domain. Distribution of natural image’s wavelet coefficients is characterized by highly kurtotic and heavytailed properties. These typical non-Gaussian statistics are commonly described by generalized Gaussian density(GGD) or α-stable distribution. However, each of the two models has its own deficiency to capture the variety and complexity of real world scenes. Considering the statistical properties of GGD and α-stable distributions respectively, we propose a hybrid statistical model of natural image’s wavelet coefficients which is better in describing the leptokurtosis and heavy tails simultaneously.Based on a clever fusion of GGD and α-stable functions, we establish the optimal parametric hybrid model, and a close-formed Kullback-Leibler divergence of the hybrid model is derived for evaluating model accuracy. Experiment results demonstrate that the proposed hybrid model is closer to the true distribution of natural image’s wavelet coefficients than the single modeling using GGD or α-stable, while is beneficial for applications such as image comparison.Further studies reveal that compared with GGD model, parameters of natural image’s multi-scale α-stable model keep invariance through scales. Consequently,Section 2 continues with image quality assessment(IQA) based on statistical comparison. Quality assessment is of central importance in numerous image processing tasks.State-of-the-art objective image quality assessment algorithms are generally devised for specific distortion types or based on training procedure of large databases. According to the availability of undistorted reference images, image quality assessment can be classified into full/reduced/no reference types(FR, RR and NR respectively).In this work, we propose a general-purpose FR/NR IQA framework for image distortions, nominated by Image Quality/Distortion Metric(IQDM). The leptokurtic and heavy-tailed behaviors of image wavelet coefficients are characterized by symmetricα-stable(SαS) density, and model parameters may be altered by the presence of distortion. This important priori knowledge of original image’s distribution is then used to gauge the distortion between degraded and reference SαS models in multi-scale wavelet sub-bands. We investigate the relationship between original and degraded parameters over scales, accordingly infer the original parameters from the degraded ones. A characteristic probability density function for SαS and its closed-form Kullback-Leibler distance are derived for FR/NR-IQDM using the model parameters.Extensive experiments indicate that the proposed FR/NR-IQDM scheme is efficacious to most common types of distortion, and leads to a highly comparable performance to the benchmarks and prevalent competitors in consistency with subjective judgements.Section 3 focuses on image noise analysis and noise parameter estimation. Noise analysis and estimation is an important premise for image denoising and many other image processing applications, and related research has drawn increasing attention and interest. Firstly, we develop a framework for estimating noise level of natural image using two important statistics: high-kurtosis and scale-invariance in transform domain. By exploring the said priors of natural image statistics in 2D discrete cosine transform(DCT) domain, we reveal the limitations of these statistics for images with highly directional edges or large smooth areas. Accordingly we derive a method integrates wavelet transform and non-directional DCT to alleviate the influence of image’s structures on its statistics, and estimate the noise variance by addressing a constrained nonlinear optimization problem. Secondly, we propose an additive noise level estimation method based on local variance distribution of image in transform domain, which is expected to provide highly efficacious and robust noise estimator for denoising algorithms. This method is not limited to any specific type of noise distribution. Thirdly,a noise-injection rectification is further devised to simulate the impact of noise-free image contents on noise estimation, and rectify the estimation error introduced by it.Simulation and comparative study demonstrate that these two noise estimation algorithms and the rectification reliably infers noise variance and is robustness over wide ranges of visual content and noise levels, while outperforms some relevant methods.Moreover, denoising results using these estimation methods verify that these works can improve the performance of representative denoiser.
Keywords/Search Tags:Generalized Gaussian distribution, α-stable distribution, hybrid model, image quality assessment, multi-scale analysis, parameter analysis of image model, Kullback-Leibler distance, noise estimation, scale invariance of kurtosis
PDF Full Text Request
Related items