Distortion Metrics Research Articles

On Function Computation With Privacy and Secrecy Constraints

In this paper, the problem of function computation with privacy and secrecy constraints is considered. The considered model consists of three legitimate nodes (i.e., two transmitters, Alice and Bob, and a fusion center that acts as the receiver) that observe correlated sources and are connected by noiseless public channels, and an eavesdropper Eve who has full access to the public channels and also has its own source observations. The fusion center would like to compute a function of the distributed sources within a prefixed distortion level under a certain distortion metric. To facilitate the function computation, Alice and Bob will send messages to the fusion center. Different from the existing setups in function computation, we assume that there is a privacy constraint on the sources at Alice and Bob. In particular, Alice and Bob would like to enable the fusion center to compute the function, but at same time, they do not want the fusion center to learn too much information about the source observations. We introduce a quantity to precisely measure the privacy leakage to the fusion center. In addition to this privacy constraint, we also have a secrecy constraint to Eve and use equivocation of sources to measure this quantity. Under this model, we study the tradeoffs among message rates, private information leakage, equivocation, and distortion. We first consider a scenario that has only one transmitter, i.e., the source at Bob is empty, and fully single-letter characterize the corresponding regions. Then, we consider the more general case and provide both outer and inner bounds on the corresponding regions.

Early termination for fast intra mode decision in depth map coding using DIS-inheritance

Depth Intra Skip (DIS) mode is a new depth intra coding tool in the 3D extension of High Efficiency Video Coding (3D-HEVC) to improve the coding efficiency of depth maps. The optimization of quadtree partitioning in HEVC is conducted on different Coding Unit (CU) sizes from 64×64 to 8×8. Although all CU sizes should be considered in partition decision, only the CUs in the optimal partition structure are coded in the final bitstream. In this paper, it is found that the CUs well predicted by DIS (i) occupy a large proportion in mode decision; (ii) have a strong consistency in its sub-CUs at the higher depth level; and (iii) tend to be coded at the lower depth level in the final bitstream. Based on these observations, this work proposes an early termination scheme for fast intra mode decision in depth maps. Early termination of CU splitting is triggered according to the consistency of DIS between different depth levels by evaluating several most probable modes in the current CUs and its four sub-CUs based on the new distortion metric. Simulation results demonstrate that the proposed algorithm could save 31.04% of depth coding time with 0.07% increase in BDBR. By integrating with other fast intra mode decision methods, the overall algorithm could provide averagely 50.08% time reduction with only 0.64% increase in BDBR.

Energy-Efficient Hadamard-Based SATD Hardware Architectures Through Calculation Reuse

The Hadamard-based Sum of Absolute Transformed Differences (SATD) is a distortion metric that correlates better with other video encoding steps than the commonly used Sum of Absolute Differences (SAD) and thus potentially improves coding efficiency. However, due to the Hadamard Transform (HT), it demands a large share of total encoding time. Meanwhile, the widespread use of embedded encoders made computational complexity and energy efficiency two relevant issues, besides the usual trade-off between the rate and distortion. Hence, the adoption of SATD during video coding requires the design of energy-efficient hardware accelerators. In this paper, we show that the SATD improves coding efficiency by up to −2.99% in BD-Rate when used in motion estimation within the HEVC Model (HM). Subsequently, as the main contribution, we combined two properties featured by the SATD to devise a novel hardware architecture based on calculation reuse. These properties allow for rearranging the architecture components to reduce the hardware size, save operations, and thus, reduce energy. The synthesis and simulation results indicate that the proposed architecture can lead to area and energy savings of about 28% and 32%, respectively, compared to the state-of-the-art architecture based on reuse solely.

Learning based Facial Image Compression with semantic fidelity metric

Surveillance and security scenarios usually require high efficient facial image compression scheme for face recognition and identification. While either traditional general image codecs or special facial image compression schemes only heuristically refine codec separately according to face verification accuracy metric. We propose a Learning based Facial Image Compression (LFIC) framework with a novel Regionally Adaptive Pooling (RAP) module whose parameters can be automatically optimized according to gradient feedback from an integrated hybrid semantic fidelity metric, including a successfully exploration to apply Generative Adversarial Network (GAN) as metric directly in image compression scheme. The experimental results verify the framework’s efficiency by demonstrating performance improvement of 71.41%, 48.28% and 52.67% Bit Rate Saving separately over JPEG2000, WebP and neural network-based codecs under the same face verification accuracy distortion metric. We also evaluate LFIC’s superior performance gain compared with latest specific facial image codecs. Visual experiments also show some interesting insight on how LFIC can automatically capture the information in critical areas based on semantic distortion metrics for optimized compression, which is quite different from the heuristic way of optimization in traditional image compression algorithms.

Real-Time Constant Objective Quality Video Coding Strategy in High Efficiency Video Coding

As video data are occupying an increasingly more significant portion of global data traffic, video communication has become an indispensable component for most multimedia applications. As an enabling technology of video communication, although video coding is well standardized, the strategy to control video codec is highly customized to applications. The consistency of video quality is being paid more attention in many emerging applications. For example, in video surveillance, the quality of video frames should be stable in order to ensure the performance of machine intelligence algorithm, such as object detection precision. In this paper, we focus on achieving constant objective reconstruction quality in the process of video coding. To achieve certain rate-distortion performance, bit rate and distortion metric are usually modeled as a function of video content and control parameters of codec. For content modeling in existing work, there is still room for improvement, including the design of more efficient content feature, the compensation for assumption about constant RD characteristics among consecutive frames, and the adjustment of Lagrangian multiplier $\lambda $ according to content property. The main contributions of this paper are: 1) a robust content adaptive model for residual bit rate modeling based on content statistics called mean absolute partial transformed difference (MAPTD); 2) a content-related header bit rate modeling; 3) preprocessing scheme for robust content feature estimation at scene change; 4) a distortion model consistent with local content; and 5) content adaptive $\lambda $ determination. The experimental results show that our constant quality control strategy can achieve superior performance compared with the state-of-the-art algorithms.

Optimal voice packet classification for enhanced VoIP over priority-enabled networks

This paper proposes a method for optimal classification of voice packets to enhance the quality of voice communications over priority-enabled networks when poor transmission conditions occur. Either high or low priority is assigned to each packet according to the relevance of its payload (voice segment) for the voice intelligibility. Then, in case of constrained networking conditions, by discarding first the voice packets of lower importance, the network always delivers those segments that most contribute to the perceptual quality. The proposed method is based on a dynamic programming optimisation algorithm that finds the optimal subset of m high priority voice segments in each utterance of size n > m. Such optimal subset minimizes the reconstruction distortion over all possible subsets with the same size m (i.e., the distortion incurred by a utterance reconstructed from only m segments). The simulation results show that the proposed method consistently achieves higher mean opinion scores (MOS) in comparison with non-selective packet drop under the same random network loss conditions, yielding better quality of experience (QoE) for the same packet loss rates (PLR). The priority classification algorithm is independent from error concealment methods and distortion metrics used in the optimisation process, which allows generalisation for diverse communication networks and applications.

Rate distortion optimization for stereoscopic video coding based on joint of binocular just-noticeable distortion and structural similarity index metric

In video coding, the Lagrange-based rate distortion optimization (RDO) plays important roles in improving coding efficiency. We proposed a method to improve the RDO of stereoscopic video coding. It uses two independent perception-based distortion metrics for the left and right views, as the distortion in the RDO. The distortion metric of the right view is to measure binocular perceptual distortion of the two views. The distortion metric of the left view is to measure two-dimensional (2-D) perceptual distortion of the left view. Both distortion metrics are developed based on the structural similarity index metric (SSIM) that is weighted by visual sensitivity. A binocular just-noticeable distortion (JND) model is proposed to measure binocular visual sensitivity for the right view. The commonly used 2-D JND models are adopted for the left view. In addition, a 1/SSIM-based rate-distortion (R-D) model is improved based the proposed distortion metrics. Based on it, the Lagrange multiplier is then derived according to the R-D theory. The experiments demonstrate that the proposed method has better performance than other perceptual RDO methods. Subjective evaluations also show that it can achieve better perceptual quality with less bit rate.

Reversible Data Hiding under Inconsistent Distortion Metrics.

Recursive code construction (RCC), based on the optimal transition probability matrix (OTPM), approaching the rate-distortion bound of reversible data hiding (RDH) has been proposed. Using the existing methods, OTPM can be effectively estimated only for a consistent distortion metric, i.e., if the host elements at different positions share the same distortion metric. However, in many applications, the distortion metrics are position dependent and should thus be inconsistent. Inconsistent distortion metrics can usually be quantified as a multi-distortion metric. In this paper, we first formulate the rate-distortion problem of RDH under a multi-distortion metric and subsequently propose a general framework to estimate the corresponding OTPM, with which RCC is extended to approach the rate-distortion bound of RDH under the multi-distortion metric. We apply the proposed framework to two examples of inconsistent distortion metrics: RDH in color image and reversible steganography. The experimental results show that the proposed method can efficiently improve upon the existing techniques.

EAD: Elastic-Net Attacks to Deep Neural Networks via Adversarial Examples

Recent studies have highlighted the vulnerability of deep neural networks (DNNs) to adversarial examples — a visually indistinguishable adversarial image can easily be crafted to cause a well-trained model to misclassify. Existing methods for crafting adversarial examples are based on L2 and L∞ distortion metrics. However, despite the fact that L1 distortion accounts for the total variation and encourages sparsity in the perturbation, little has been developed for crafting L1-based adversarial examples. In this paper, we formulate the process of attacking DNNs via adversarial examples as an elastic-net regularized optimization problem. Our elastic-net attacks to DNNs (EAD) feature L1-oriented adversarial examples and include the state-of-the-art L2 attack as a special case. Experimental results on MNIST, CIFAR10 and ImageNet show that EAD can yield a distinct set of adversarial examples with small L1 distortion and attains similar attack performance to the state-of-the-art methods in different attack scenarios. More importantly, EAD leads to improved attack transferability and complements adversarial training for DNNs, suggesting novel insights on leveraging L1 distortion in adversarial machine learning and security implications of DNNs.

Characteristics of unsteady total pressure distortion for a complex aero-engine intake duct

Some types of aero-engine intake systems are susceptible to the generation of secondary flows with high levels of total pressure fluctuations. The resulting peak distortion events may exceed the tolerance level of a given engine, leading to handling problems or to compressor surge. Previous work used distortion descriptors for the assessment of intake-engine compatibility to characterise modestly curved intakes where most of the self-generated time-dependent distortion was typically found to be dominated by stochastic events. This work investigates the time-dependent total pressure distortion at the exit of two high off-set diffusing S-duct intakes with the aim of establishing whether this classical approach, or similar, could be applied in these instances. The assessment of joint probability maps for time dependent radial and circumferential distortion metrics demonstrated that local ring-based distortion descriptors are more appropriate to characterise peak events. Extreme Value Theory (EVT) was applied to predict the peak distortion levels that could occur for a test time beyond the experimental data set available. Systematic assessments of model sensitivities to the de-clustering frequency, the number of exceedances and sample time length were used to extend the EVT application to local distortion descriptors and to provide guidelines on its usage.

QVZ: lossy compression of quality values.

MOTIVATION Recent advancements in sequencing technology have led to a drastic reduction in the cost of sequencing a genome. This has generated an unprecedented amount of genomic data that must be stored, processed and transmitted. To facilitate this effort, we propose a new lossy compressor for the quality values presented in genomic data files (e.g. FASTQ and SAM files), which comprise roughly half of the storage space (in the uncompressed domain). Lossy compression allows for compression of data beyond its lossless limit. RESULTS The proposed algorithm QVZ exhibits better rate-distortion performance than the previously proposed algorithms, for several distortion metrics and for the lossless case. Moreover, it allows the user to define any quasi-convex distortion function to be minimized, a feature not supported by the previous algorithms. Finally, we show that QVZ-compressed data exhibit better performance in the genotyping than data compressed with previously proposed algorithms, in the sense that for a similar rate, a genotyping closer to that achieved with the original quality values is obtained. AVAILABILITY AND IMPLEMENTATION QVZ is written in C and can be downloaded from https://github.com/mikelhernaez/qvz. CONTACT mhernaez@stanford.edu or gmalysa@stanford.edu or iochoa@stanford.edu SUPPLEMENTARY INFORMATION Supplementary data are available at Bioinformatics online.

Assessment methods for unsteady flow distortion in aero-engine intakes

Peak events of unsteady total pressure and swirl distortion generated within S-duct intakes can affect the engine stability, even when within acceptable mean distortion levels. Even though the swirl distortion descriptors have been evaluated in S-duct intakes, the associated flow field pattern has not been reported in detail. This is of importance since engine tolerance to distortion is usually tested with representative patterns from intake tests replicated with steady distortion generators. Despite its importance in intake/engine compatibility assessments, the spectral characteristics of the distortion descriptors and the relationship between peak unsteady swirl and both radial and circumferential total pressure distortion has not been assessed previously. The peak distortion data is typically low-pass filtered at a frequency associated with the minimum response time of the engine. However the engine design is not always known a priori in intakes investigations and a standard approach to reporting peak distortion data is needed. In addition, expensive and time-consuming tests are usually required to capture representative extreme distortion levels. This work presents a range of analyses based on Delayed Detached-Eddy Simulation and Stereo Particle Image Velocimetry data at the outlet of a representative S-duct intake to assess these aspects of the unsteady flow distortion. The distorted pattern associated with different swirl distortion metrics is identified based on a conditional averaging technique, which indicates that the most intense swirl events are associated with a single rotating structure. The main frequencies of the flow distortion descriptors are found to lie within the range in which the engine stability may be compromised. The peak total pressure and swirl distortion events are found to be not synchronous, which highlights the need to assess both types of distortion. Peak swirl and total-pressure distortion data is reported as a function of its associated time scale in a more general way that can be used in the assessment of intake unsteady flow distortion. Extreme Value Theory has been applied to predict peak distortion values beyond those measured in the available dataset, and whose measurement would otherwise require testing times two orders of magnitude longer than those typically considered.

Experimental measurement of residual stress and distortion in additively manufactured stainless steel components with various dimensions

Disk-shaped 316L stainless steel parts with various diameters and heights were additively manufactured using a direct metal laser sintering (DMLS) technique. Neutron diffraction was used to profile the residual stresses in the samples before and after removal of the build plate and support structures. Moreover, distortion level of the parts before and after the removal was quantified using a coordinate measuring machine (CMM). Large tensile in-plane stresses (up to ≈ 400MPa) were measured near the as-built disk top surfaces, where the stress magnitude decreased from the disk center to the edges. The stress gradient was steeper for the disks with smaller diameters and heights. Following the removal of the build plate and support structures, the magnitude of the in-plane residual stresses decreased dramatically (up to 330MPa) whereas the axial stress magnitude did not change significantly. The stress relaxation caused the disks to distort, where the distortion metric was higher for the disks with smaller diameters and heights. The distribution of the residual stresses revealed a marked breakdown of self-similarity in their distribution even comparing disk-shaped samples that were fabricated under identical printing parameters; the stress field profiles were not linearly scaled as a function of height and diameter.

Enabling Reliable Change Detection for Independently Compressed SAR Images

This paper develops distortion metrics for compressed synthetic aperture radar (SAR) imagery from change detection test statistics. These metrics are used to predict lossy image compression’s impact on change detection performance. The metrics do not require the intended change detection comparison image to provide these benefits. An SAR compression system leveraging the distortion metrics is proposed. The system generates a bad-pixel mask highlighting potential false alarms that are generated due to compression and are subsequently discarded in the change detection process. The proposed system’s performance is demonstrated through noncoherent change detection analysis after JPEG2000 and JPEG image compression. Similarly, a coherent change detection system is evaluated after JPEG2000 image compression. For noncoherent change detection at large compression ratios (CRs) using JPEG2000, the proposed system provides a 33% reduction in false alarms at a 0.1 probability of detection as well as the ability to maintain near-distortionless false alarm rates across a wide range of CRs. At a 0.1 probability of detection for coherent change detection, the system provides a 37% reduction in false alarms at modest CRs. The coherent change detection system is also demonstrated to maintain low false alarm rates across a range of CRs.

Pressure Flowfield and Inlet Flow Distortion Metrics Reconstruction from Velocity Data

Complex engine intakes are susceptible to unsteady flow distortions that may compromise the propulsion system operability. Hence, the need for high spatial and temporal resolution flow information is essential to aid the development of distortion-tolerant closely coupled propulsion systems. Stereoscopic particle image velocimetry methods have been successfully applied to these flows, offering synchronous velocity datasets of high spatial resolution across the aerodynamic interface plane. However, the total pressure distortion measurements are still typically provided by low bandwidth intrusive total pressure rakes of low spatial resolution, which results in limited characterization of the total pressure distortion. This limitation can potentially be addressed by pressure field reconstruction from non-intrusive high resolution velocity data. A range of reconstruction methods is assessed based on representative data from steady and unsteady computational simulations of an S-duct configuration. In addition to the reconstructed total pressure field, the impact on the key distortion metrics is assessed. The effect of the Mach number is considered. Overall, the reconstruction methods show that the distortion metrics can be determined with sufficient accuracy to indicate that there is a potential benefit from exploiting high resolution velocity measurements in evaluating total pressure distortion.

Complex Aeroengine Intake Ducts and Dynamic Distortion

For many embedded engine systems, the intake duct geometry introduces flow distortion and unsteadiness, which must be understood when designing the turbomachinery components. The aim of this work is to investigate the capabilities of modern computational methods for these types of complex flows, to study the unsteady characteristics of the flowfield, and to explore the use of proper orthogonal decomposition methods to understand the nature of the unsteady flow distortion. The unsteady flows for a range of S-duct configurations have been simulated using a delayed detached-eddy simulation method. Analysis of the conventional distortion criteria highlights the main sensitivities to the S-duct configuration and quantifies the unsteady range of these parameters. The unsteady flowfield shows signature regions of unsteadiness, which are postulated to be related to the classical secondary flows as well as to the streamwise flow separation. A proper orthogonal decomposition of the total pressure field at the duct exit identifies the underpinning flow modes, which are associated with the overall total pressure unsteadiness distributions. Overall, the unsteady distortion metrics are not found to be solely linked to a particular proper orthogonal decomposition mode, but are dependent on a wider range of modes.

Reversible data hiding for 3D mesh models with three-dimensional prediction-error histogram modification

This study proposes a reversible data hiding (RDH) algorithm for 3D mesh models based on the optimal three-dimensional prediction-error histogram (PEH) modification with recursive construction coding (RCC). Firstly, we design a double-layered prediction scheme to divide all the vertices of 3D mesh model into the “embedded” set and the “referenced” set, according to the odd-even property of indices into the vertex list. Thanks to the geometrical similarity among neighboring vertices, we obtain the prediction errors (PEs) with a sharp histogram. Then we combine every three adjacent PEs into one prediction-error triplet (PET) and construct the three-dimensional PEH with smaller entropy than one-dimensional PEH by utilizing the correlation among PEs. Next we project the three-dimensional PEH into one-dimensional space for scalar PET sequence which is suitable for using RCC. And also, we define the distortion metrics for 3D mesh models, by which we can estimate the optimal probability transition matrix (OTPM) indicating the optimal PEH modification manner. After that, we modify the PET sequence and embed data by RCC according to OTPM. The experimental results show that our method is superior to two state-of-the-art spatial-domain RDH algorithms for 3D mesh models a lot.

A Secure Steganographic Algorithm Based on Frequency Domain for the Transmission of Hidden Information

This contribution proposes a novel steganographic method based on the compression standard according to the Joint Photographic Expert Group and an Entropy Thresholding technique. The steganographic algorithm uses one public key and one private key to generate a binary sequence of pseudorandom numbers that indicate where the elements of the binary sequence of a secret message will be inserted. The insertion takes eventually place at the first seven AC coefficients in the transformed DCT domain. Before the insertion of the message the image undergoes several transformations. After the insertion the inverse transformations are applied in reverse order to the original transformations. The insertion itself takes only place if an entropy threshold of the corresponding block is satisfied and if the pseudorandom number indicates to do so. The experimental work on the validation of the algorithm consists of the calculation of the peak signal-to-noise ratio (PSNR), the difference and correlation distortion metrics, the histogram analysis, and the relative entropy, comparing the same characteristics for the cover and stego image. The proposed algorithm improves the level of imperceptibility analyzed through the PSNR values. A steganalysis experiment shows that the proposed algorithm is highly resistant against the Chi-square attack.

Just-Noticeable Difference-Based Perceptual Optimization for JPEG Compression

The Quantization table in JPEG, which specifies the quantization scale for each discrete cosine transform (DCT) coefficient, plays an important role in image codec optimization. However, the generic quantization table design that is based on the characteristics of human visual system (HVS) cannot adapt to the variations of image content. In this letter, we propose a just-noticeable difference (JND) based quantization table derivation method for JPEG by optimizing the rate-distortion costs for all the frequency bands. To achieve better perceptual quality, the DCT domain JND-based distortion metric is utilized to model the stair distortion perceived by HVS. The rate-distortion cost for each band is derived by estimating the rate with the first-order entropy of quantized coefficients. Subsequently, the optimal quantization table is obtained by minimizing the total rate-distortion costs of all the bands. Extensive experimental results show that the quantization table generated by the proposed method achieves significant bit-rate savings compared with JPEG recommended quantization table and specifically developed quantization tables in terms of both objective and subjective evaluations.

On the Rate-Distortion Function for Binary Source Coding With Side Information

We present an in-depth analysis of the problem of lossy compression of binary sources in the presence of correlated side information, where the correlation is given by a generic binary asymmetric channel and the Hamming distance is the distortion metric. Our analysis is motivated by systematic rate-distortion gains observed when applying asymmetric correlation models in Wyner-Ziv video coding. First, we derive for the first time the rate-distortion function for conventional predictive coding in the binary-asymmetric-correlation-channel scenario. Second, we propose a new bound for the case where the side information is only available at the decoder—Wyner-Ziv coding. We conjecture this bound to be tight. We show that the maximum rate needed to encode as well as the maximum rate-loss of Wyner-Ziv coding relative to predictive coding corresponds to uniform sources and symmetric correlations. Importantly, we show that the upper bound on the rate-loss established by Zamir is not tight and that the maximum value is actually significantly lower. Moreover, we prove that the only binary correlation channel that incurs no rate-loss for Wyner-Ziv coding compared with predictive coding is the Z-channel. Finally, we complement our analysis with new compression performance results obtained with our state-of-the-art Wyner-Ziv video coding system.