DCT-based Filter Research Articles

Analysis of the potential efficiency of post-filtering noisy images after lossy compression

An increase in the number of images and their average size is the general trend nowadays. This increase leads to certain problems with data storage and transfer via communication lines. A common way to solve this problem is to apply lossy compression that provides sufficiently larger compression ratios compared to lossless compression approaches. However, lossy compression has several peculiarities, especially if a compressed image is corrupted by quite intensive noise. First, a specific noise-filtering effect is observed. Second, an optimal operational point (OOP) might exist where the quality of a compressed image is closer to the corresponding noise-free image than the quality of the original image according to a chosen quality metric. In this case, it is worth compressing this image in the OOP or its closest neighborhood. These peculiarities have been earlier studied and their positive impact on image quality improvement has been demonstrated. Filtering of noisy images due to lossy compression is not perfect. Because of this, it is worth checking can additional quality improvement be reached using such an approach as post-filtering. In this study, we attempt to answer the questions: “is it worth to post-filter an image after lossy compression, especially in OOP’s neighborhood? And what benefit can it bring in the sense of image quality?”. The study is carried out for better portable graphics (BPG) coder and the DCT-based filter focusing mainly on one-component (grayscale) images. The quality of images is characterized by several metrics such as PSNR, PSNR-HVS-M, and FSIM. Possible image quality increasing via post-filtering is demonstrated and the recommendations for filter parameter setting are given.

Feature-filter: Detecting adversarial examples by filtering out recessive features

Deep neural networks (DNNs) have achieved state-of-the-art performance in numerous tasks involving complex analysis of raw data, such as self-driving systems and biometric recognition systems. However, recent works have shown that DNNs are under threat from adversarial example attacks. The adversary can easily change the outputs of DNNs by adding small well-designed perturbations to inputs. Adversarial example detection is fundamental for robust DNN-based services. From a human-centric perspective, this paper divides image features into dominant features comprehensible to humans and recessive features incomprehensible to humans yet exploited by DNNs. Based on this perspective, the paper proposes a new viewpoint that imperceptible adversarial examples are the product of recessive features misleading neural networks, and that the adversarial attack enriches these recessive features. The imperceptibility of the adversarial examples indicates that the perturbations enrich recessive features but hardly affect dominant features. Therefore, adversarial examples are sensitive to filtering out recessive features, while benign examples are immune to such operations. Inspired by this idea, we propose a label-only adversarial detector that is referred to as a feature-filter. The feature-filter utilizes the discrete cosine transform (DCT) to approximately separate recessive features from dominant features and obtain a filtered image. A comprehensive user study demonstrates that the DCT-based filter can reliably filter out recessive features from the test image. By comparing only the DNN’s prediction labels on the input and its filtered version, the feature-filter can detect imperceptible adversarial examples in real time with high accuracy and few false-positives.

Reduction of Spatially Correlated Speckle in Textured SAR Images

Synthetic aperture radars (SARs) provide a lot of images that can be used for numerous applications. A problem with acquired images is that they are corrupted by speckle which is a noise-like phenomenon with multiplicative nature. In addition, speckle is non-Gaussian and it is often spatially correlated. A typical task in SAR image processing is despeckling and many methods have been already proposed. However, most of them do not take noise spatial correlation into account during denoising. In this paper, we show how this can be done in despeckling based on discrete cosine transform. The use of frequency-dependent thresholds leads to sufficient improvement of denoising efficiency in terms of visual quality metrics. Moreover, we consider quite complex structure texture images for which noise removal is usually problematic and can lead to information loss. Comparison to the well-known local statistic Lee and Frost filters, extended DCT-based filter is carried out for different remote sensing systems including Sentinel-1 and Sentinel-2.

Prediction of performance of 2D DCT-based filter and adaptive selection of its parameters

To enhance images, one often has to apply a filtering operation (denoising). However, there are several issues within the denoising. One of them is that sometimes denoising can be not efficient. Another issue regards a selection of an appropriate filter and setting of its parameters. As a particular case, we consider a 2D DCTbased filter with 8x8 pixel fully overlapping blocks where one of the parameters is a proportionality factor (PF) used in the threshold setup. We show that a performance of the considered filter in the sense of standard PSNR and visual quality metric PSNR-HVS-M can be predicted before applying image filtering procedure. This prediction is sufficiently faster than the denoising itself and accurate enough. We demonstrate that, having DCT statistics in a limited number of image blocks, such a prediction can be done for several values of PF. This allows deciding is it worth applying filtering to an image at hand. If the denoising is desired, it is also possible to select the PF optimal value for the considered image and noise intensity. Such a procedure, in some cases, can result in improvement of output PSNR or PSNR-HVS-M by up to 1 dB in comparison to the default parameters setup.

ANALYSIS OF OPPORTUNITIES TO IMPROVE IMAGE DENOISING EFFICIENCY FOR DCT-BASED FILTER

The subject matter of the paper is the process of image filtering. The goal is to provide high efficiency of denoising according to metrics that are more adequate to human vision system than traditional criteria as mean square error or peak signal-to-noise ratio. The tasks to be solved are the following: to carry out analysis of denoising efficiency using visual quality metric, to determine optimal settings of DCT-based filter depending upon image and noise properties, to propose a method for setting a global threshold adaptively (in quasi-optimal manner) based on preliminary analysis of image and noise properties. The following results have been obtained: 1) optimal values of filter parameters depend upon many factors including image complexity and noise intensity, 2) optimal values also depend on optimization criterion (or metric) used to characterize filter performance; 3) optimal values of parameter β that determines the threshold can considerably differ from 2.6 which is traditionally recommended; 4) this opens opportunities for improving image denoising efficiency; 5) one of this opportunities consists in preliminary analysis of image and noise properties with setting the threshold value according to the obtained dependences. Conclusions: 1) the filter performance can be sufficiently improved due to the proposed adaptive procedure; 2) this happens if either noise is intensive and image has a simple structure or if noise is not too intensive and image has a complex structure; 3) the proposed adaptive procedure requires a very small amount of additional computations for calculating input parameter and can be realized by 60 or more times faster than filtering itself; 4) the adaptive procedure slightly differs depending upon a metric used as performance criterion.

PERFORMANCE COMPARISON OF DENOISING METHODS FOR HISTORICAL DOCUMENTS

Image denoising plays an important role in image processing. It is also part of the pre-processing technique in a binarization complete procedure that consists of pre-processing, thresholding, and post-processing. Our previous research has confirmed that the Discrete Cosine Transform (DCT)-based filtering as the new pre-processing process improved the performance of binarization output in terms of recall and precision. This research compares three classical denoising methods; Gaussian, mean, and median filtering with the DCT-based filtering. The noisy ancient document images are filtered using those classical filtering methods. The outputs of this process are used as the input for Otsu, Niblack, Sauvola and NICK binarization methods. Then the resulted binary images of the three classical methods are compared with those of DCT-based filtering. The performance of all denoising algorithms is evaluated by calculating recall and precision of the resulted binary images. The result of this research is that the DCT based filtering resulted in the highest recall and precision as compared to the other methods.

Efficiency of texture image enhancement by DCT-based filtering

Textures or high-detailed structures as well as image object shapes contain information that is widely exploited in pattern recognition and image classification. Noise can deteriorate these features and has to be removed. In this paper, we consider the influence of textural properties on efficiency of image enhancement by noise suppression for the posterior treatment. Among possible variants of denoising, filters based on discrete cosine transform known to be effective in removing additive white Gaussian noise are considered. It is shown that noise removal in texture images using the considered techniques can distort fine texture details. To detect such situations and to avoid texture degradation due to filtering, filtering efficiency predictors, including neural network based predictor, applicable to a wide class of images are proposed. These predictors use simple statistical parameters to estimate performance of the considered filters. Image enhancement is analysed in terms of both standard criteria and metrics of image visual quality for various scenarios of texture roughness and noise characteristics. The discrete cosine transform based filters are compared to several counterparts. Problems of noise removal in texture images are demonstrated for all of them. A special case of spatially correlated noise is considered as well. Potential efficiency of filtering is analysed for both studied noise models. It is shown that studied filters are close to the potential limits.

AN IMPROVED PREDICTION OF DCT-BASED IMAGE FILTERS EFFICIENCY USING REGRESSION ANALYSIS

Efficiency of DCT-based filters for a wide-class of images is investigated. The study is carried out for additive white Gaussian noise (AWGN) case with several intensity levels. Local DCT-based filter is used as basic denoising technique. Nonlocal BM3D filter known as the state-of-the-art technique for AWGN removal is also exploited. A precise prediction method of denoising efficiency for several quality metrics is proposed. It is shown that statistics of DCT coefficients provides useful information. Regression models for analyzed filters and metrics are presented. The obtained dependence approximations of quality metrics on DCT statistics have high goodness of fit. One-parameter and multi-parameter fitting cases are considered. The most valuable DCT statistics are found.

Classification of interstitial lung disease patterns using local DCT features and random forest.

Over the last decade, a plethora of computer-aided diagnosis (CAD) systems have been proposed aiming to improve the accuracy of the physicians in the diagnosis of interstitial lung diseases (ILD). In this study, we propose a scheme for the classification of HRCT image patches with ILD abnormalities as a basic component towards the quantification of the various ILD patterns in the lung. The feature extraction method relies on local spectral analysis using a DCT-based filter bank. After convolving the image with the filter bank, q-quantiles are computed for describing the distribution of local frequencies that characterize image texture. Then, the gray-level histogram values of the original image are added forming the final feature vector. The classification of the already described patches is done by a random forest (RF) classifier. The experimental results prove the superior performance and efficiency of the proposed approach compared against the state-of-the-art.

Analysis of classification accuracy for pre-filtered multichannel remote sensing data

Noise is one of the main factors degrading the quality of original multichannel remote sensing data and its presence influences classification efficiency, object detection, etc. Thus, pre-filtering is often used to remove noise and improve the solving of final tasks of multichannel remote sensing. Recent studies indicate that a classical model of additive noise is not adequate enough for images formed by modern multichannel sensors operating in visible and infrared bands. However, this fact is often ignored by researchers designing noise removal methods and algorithms. Because of this, we focus on the classification of multichannel remote sensing images in the case of signal-dependent noise present in component images. Three approaches to filtering of multichannel images for the considered noise model are analysed, all based on discrete cosine transform in blocks. The study is carried out not only in terms of conventional efficiency metrics used in filtering (MSE) but also in terms of multichannel data classification accuracy (probability of correct classification, confusion matrix). The proposed classification system combines the pre-processing stage where a DCT-based filter processes the blocks of the multichannel remote sensing image and the classification stage. Two modern classifiers are employed, radial basis function neural network and support vector machines. Simulations are carried out for three-channel image of Landsat TM sensor. Different cases of learning are considered: using noise-free samples of the test multichannel image, the noisy multichannel image and the pre-filtered one. It is shown that the use of the pre-filtered image for training produces better classification in comparison to the case of learning for the noisy image. It is demonstrated that the best results for both groups of quantitative criteria are provided if a proposed 3D discrete cosine transform filter equipped by variance stabilizing transform is applied. The classification results obtained for data pre-filtered in different ways are in agreement for both considered classifiers. Comparison of classifier performance is carried out as well. The radial basis neural network classifier is less sensitive to noise in original images, but after pre-filtering the performance of both classifiers is approximately the same.

Wiener discrete cosine transform-based image filtering

A classical problem of additive white (spatially uncorrelated) Gaussian noise suppression in grayscale images is considered. The main attention is paid to discrete cosine transform (DCT)-based denoising, in particular, to image processing in blocks of a limited size. The efficiency of DCT-based image filtering with hard thresholding is studied for different sizes of overlapped blocks. A multiscale approach that aggregates the outputs of DCT filters having different overlapped block sizes is proposed. Later, a two-stage denoising procedure that presumes the use of the multiscale DCT-based filtering with hard thresholding at the first stage and a multiscale Wiener DCT-based filtering at the second stage is proposed and tested. The efficiency of the proposed multiscale DCT-based filtering is compared to the state-of-the-art block-matching and three-dimensional filter. Next, the potentially reachable multiscale filtering efficiency in terms of output mean square error (MSE) is studied. The obtained results are of the same order as those obtained by Chatterjee's approach based on nonlocal patch processing. It is shown that the ideal Wiener DCT-based filter potential is usually higher when noise variance is high.