Median filtering forensics based on discriminative multi-scale sparse coding

The sparse coding method is formulated as an information theoretic optimization problem. The rate distortion theory leads to an objective functional which can be interpreted as an information theoretic formulation of the sparse coding. Viewing as an entropy minimization problem, the rate distortion theory and consequently the sparse coding are extended to discriminative variants. As a concrete example of this information theoretic sparse coding, a discriminative non-linear sparse coding algorithm with neural networks is proposed. Experimental results of gender classification by face images show that the discriminative sparse coding is more robust to noise, compared to the conventional method which directly uses images as inputs to a linear support vector machine.

This paper addresses the problem of energy disaggregation/non-intrusive load monitoring. It introduces a new method based on the transform learning formulation. Several recent techniques, such as discriminative sparse coding, powerlet disaggregation, and deep sparse coding, are based on the synthesis dictionary learning/sparse coding approach; ours is based on its analysis equivalent. The theoretical advantage of analysis dictionary compared to its synthesis counterpart is that the former can learn from fewer training s--this has implications in reducing the cost of energy disaggregation. Experiments have been carried out on two benchmark data sets-REDD and Dataport (Pecan Street). Comparison has been done with factorial HMM, discriminative sparse coding, powerlet disaggregation, and deep sparse coding. In the low training data regime, our method always excels over the others.

Recent research has shown the initial success of sparse coding (Sc) in solving many computer vision tasks. Motivated by the fact that kernel trick can capture the nonlinear similarity of features, which helps in finding a sparse representation of nonlinear features, we propose kernel sparse representation (KSR). Essentially, KSR is a sparse coding technique in a high dimensional feature space mapped by an implicit mapping function. We apply KSR to feature coding in image classification, face recognition, and kernel matrix approximation. More specifically, by incorporating KSR into spatial pyramid matching (SPM), we develop KSRSPM, which achieves a good performance for image classification. Moreover, KSR-based feature coding can be shown as a generalization of efficient match kernel and an extension of Sc-based SPM. We further show that our proposed KSR using a histogram intersection kernel (HIK) can be considered a soft assignment extension of HIK-based feature quantization in the feature coding process. Besides feature coding, comparing with sparse coding, KSR can learn more discriminative sparse codes and achieve higher accuracy for face recognition. Moreover, KSR can also be applied to kernel matrix approximation in large scale learning tasks, and it demonstrates its robustness to kernel matrix approximation, especially when a small fraction of the data is used. Extensive experimental results demonstrate promising results of KSR in image classification, face recognition, and kernel matrix approximation. All these applications prove the effectiveness of KSR in computer vision and machine learning tasks.

Abnormalities in the rhythmic electromechanical contractions of the heart results in cardiac arrhythmias. When these abnormalities rise from the ventricles of the heart, they are classified as ventricular arrhythmias. The two major types of ventricular arrhythmias are ventricular tachycardia (VT) and ventricular fibrillation (VF). Ventricular fibrillation is the most dangerous among the two arrhythmias, that usually leads to sudden cardiac death if not treated immediately. Annually about 40,000 sudden cardiac deaths are reported in Canada. Due to high mortality rate and serious impact on quality of life, researchers have been focusing on characterizing ventricular arrhythmias that may lead to delivering optimized treatment options in improving the survival rates. In this thesis two major types of ventricular arrhythmias were analyzed and quantified by performing discriminative sparse coding analysis called label consistent K-SVD using time frequency dictionaries that are well localized in time and frequency domains. The analyzed signals were 670 ECG ventricular arrhythmia segments from 33 patients extracted from the Malignant Ventricular Ectopy and Creighton University Tachy-Arrhythmia databases. Using the LCKSVD dictionary learning approach, an overall maximum classification accuracy of 73.3% was achieved with a hybrid optimized wavelet dictionary. Based on the comparative analysis, the trained (learned) dictionaries yielded better performance than the untrained dictionaries. The results indicate that discriminative sparse coding approach has greater potential in extracting signal adaptive and morphologically discriminative time-frequency structures in studying ventricular arrhythmias.

In digital image forensics, it is generally accepted that intentional manipulations of the image content are most critical and hence numerous forensic methods focus on the detection of such ‘malicious ’ post-processing. However, it is also beneficial to know as much as possible about the general processing history of an image, including content-preserving operations, since they can affect the reliability of forensic methods in various ways. In this paper, we present a simple yet effective technique to detect median filtering in digital images—a widely used denoising and smoothing operator. As a great variety of forensic methods relies on some kind of a linearity assumption, a detection of non-linear median filtering is of particular interest. The effectiveness of our method is backed with experimental evidence on a large image database.

Tampering detection has been increasingly attracting attention in the field of digital forensics. As a popular nonlinear smoothing filter, median filtering is often used as a post-processing operation after image forgeries such as copy-paste forgery (including copy-move and image splicing), which is of particular interest to researchers. To implement the blind detection of median filtering, this paper proposes a novel approach based on a frequency-domain feature coined the annular accumulated points (AAP). Experimental results obtained on widely used databases, which consists of various real-world photos, show that the proposed method achieves outstanding performance in distinguishing median-filtered images from original images or images that have undergone other types of manipulations, especially in the scenarios of low resolution and JPEG compression with a low quality factor. Moreover, our approach remains reliable even when the feature dimension decreases to 5, which is significant to save the computing time required for classification, demonstrating its great advantage to be applied in real-time processing of big multimedia data.

In feature learning field, many methods are inspired by advances in neuroscience. Among them, neural network and sparse coding have been broadly studied. Predictive sparse decomposition (PSD) is a practical variant of these two methods. It trains a neural network to estimate the sparse codes. After training, the neural network is fine-tuned to achieve higher performance on object recognition tasks. It is widely believed that introducing discriminative information can make the features more useful for classification task. Hence, in this work, we propose applying the task-driven dictionary learning framework to the PSD and demonstrate that this new model can be optimized by the stochastic gradient descent (SGD) algorithm. Before our work, the semi-supervised auto-encoder framework has already been proposed to guide neural network to extract discriminative representations. But it does not improve the classification performance of neural network. In the experiments, we compare the proposed method with the semi-supervised auto-encoder method. The performance of PSD is used as the baseline for these two methods. On the MNIST and USPS datasets, our method can generate more discriminative and predictable sparse codes than other methods. Furthermore, the recognition accuracy of neural network can be improved.

Recently, the detection of median filtering (MF), which is a popular nonlinear denoising manipulation, has attracted extensive attention from researchers. Several detectors with satisfying performance have been developed, while most of them need to train proper classifiers and their performance may be degraded under JPEG compression. In this paper, a training-free MF detector with single-dimensional feature is proposed based on the difference of frequency residuals, which can solve the detection issue of median filtering images under JPEG post-processing. It is designed relying on the fact that when an image is median filtered over and over again, the frequency residual obtained from continuous two images monotonically decreases. The difference between the frequency residuals obtained from the first MF and the second MF is pretty large in an unfiltered test image, while it is relatively small if the test image is a median filtered one. Thus, the unfiltered and the median filtered images are distinguishable. Furthermore, a novel strategy combining unsharp masking (USM) sharpening is implemented to suppress the effect of image content and find a universal threshold which is utilized to classify two types of images. Experimental results show that the proposed method outperforms some state-of-the-art methods at the condition of a low false alarm rate, especially when the test images are in low quality and low resolution.

We mainly propose a robust Embedded Locality-Constrained Label Consistent Dictionary Learning (ELC2DL) framework for discriminative classification. ELC2DL improves the representation and classification performance by performing DL in the noise-removed sparse embedding space, since most real data often contains noise and performing DL over noisy data for reconstruction may decrease performance potentially. To reduce the noise in data, our model computes a sparse projection jointly for noise reduction and then uses the noise-removed data for DL. By incorporating a noise-reduction term with a discriminative locality-constrained label consistent term that associates the label information with each dictionary atom to preserve local structure of training data, a noise-reduction projection, an over-complete dictionary and discriminative sparse codes are obtained jointly. Simulations on several image databases show that our algorithm can deliver enhanced performance over other state-of-the-arts.

In multimedia forensics, exposing an image’s processing history draws much attention. Median filtering is a popular noise removal tool, which has been used as a popular anti-forensics tool recently. An image is usually saved in a compressed format such as the JPEG format. While the detection of median filtering from a JPEG compressed image is difficult because typical filter characteristics are suppressed by JPEG quantization and block artifacts. In this paper, we introduce a novel forensic trace – median filter residual. Median filtering is first applied on a test image, and the difference between the initial image and the filtered output image is called the median filter residual. The filtered residual is used as the forensic fingerprint. Thus, the interference from the image edge and texture which is regarded as a major limitation of the existing forensic methods can be reduced. Experimental results on a large image database show that the proposed method is very robust to JPEG post- compression, and achieves much better performance than the existing state-of-the-art works.KeywordsDigital image ForensicsMedian filter residualTransition probability

In recent years, sparse coding via dictionary learning has been widely used in many applications for exploiting sparsity patterns of data. For classification, useful sparsity patterns should have discrimination, which cannot be well achieved by standard sparse coding techniques. In this paper, we investigate structured sparse coding for obtaining discriminative class-specific group sparsity patterns in the context of classification. A structured dictionary learning approach for sparse coding is proposed by considering the \(\ell _{2,0}\) norm on each class of data. An efficient numerical algorithm with global convergence is developed for solving the related challenging \(\ell _{2,0}\) minimization problem. The learned dictionary is decomposed into class-specific dictionaries for the classification that is done according to the minimum reconstruction error among all the classes. For evaluation, the proposed method was applied to classifying both the synthetic data and real-world data. The experiments show the competitive performance of the proposed method in comparison with several existing discriminative sparse coding methods.

Digital images are one of the primary modern media for information interchange. However, digital images are vulnerable to interception and manipulation due to the wide availability of image editing software tools. Filtering forgery detection and splicing detection are two of the most important problems in digital image forensics. In particular, the primary challenge for the filtering forgery detection problem is that typically the techniques effective for nonlinear filtering (e.g. median filtering) detection are quite ineffective for linear filtering detection, and vice versa. In this paper, we have used Discrete Cosine Transform Residual features to train a Support Vector Machine classifier, and have demonstrated its effectiveness for both linear and non-linear filtering (specifically, Median Filtering) detection and filter classification, as well as re-compression based splicing detection in JPEG images. We have also theoretically justified the choice of the abovementioned feature set for both type of forgeries. Our technique outperforms the state-of-the-art forensic techniques for filtering detection, filter classification and re-compression based splicing detection, when applied on a set of standard benchmark images.

In recent years, the median filtering operation is increasingly used as a commonly used image post-processing method to hide the traces of other tampering operations. The detection of image tampering with median filtering has gradually become an important branch in the field of digital image forensics. However, the existing median filter detector field lacks a high-precision method to detect whether the image has undergone a median filter operation in terms of low-resolution images compressed with low quality factors. This paper combines the characteristics of median filter and proposes a median filter detection method based on convolutional neural network (CNN), which can automatically extract features directly from the image and build a high-precision detector. The first layer of the CNN Network in this paper adopts median filtered residuals (MFRs) of original and median filtered images. Then, the hierarchical representation is learned by alternating convolutional layers and pooling layers, and multiple features are extracted for further classification. At the same time, based on the detection of the median filter algorithm by the convolutional neural network, in order to further increase the accuracy of the algorithm detection, this paper supplements the method of wavelet high-frequency coefficients of the median filtered residual to extract the feature vector. Finally, extensive experiments show that algorithm can obtain a better performance by CNN and wavelet high-frequency coefficients algorithm, which is of importance in practical applications.

Attenuation of random noise has been extensively covered by f-x predictive deconvolution, median filtering and decomposition into principal components. We suggest instead using sparse code shrinkage and we propose a method to obtain the necessary noise-free realization of the data. This realization, that is in fact noise-attenuated and that aims at having the same statistical properties as the seismic signal, allows us to derive shrinkage functions based on the noise-free sparse code statistics, i.e., that are specifically designed for the input data. This method, which is tested on both synthetic and real marine data, turns out to be efficient and promising in random noise attenuation.

Sparse coding is a method for finding a neural network representation of multidimensional data in which each of the components of the representation is rarely ignorantly active at the same time. The representation is closely related to independent component analysis (ICA). In this paper, we introduced the basic principle of ICA and have investigated the capabilities of sparse coding shrinkage in the field of image denoising. We have also performed practical implementation of sparse code shrinkage (SCS) and applied to the image denoising. We have seen that SCS outperforms basic denoising methods such as wiener filtering, median filtering and independent component analysis (ICA) applied to image denoising.