W. David Pan

On fast and accurate block-based motion estimation algorithms using particle swarm optimization

Both fast and accurate block-matching algorithms are critical to efficient compression of video frames using motion estimation and compensation. While the particle swarm optimization approach holds the promise of alleviating the local optima problem suffered typically by existing very fast block matching methods, motion estimation algorithms based on particle swarm optimization in the literature appear to be either much slower than some leading fast block-matching methods for a given accuracy of motion estimation, or less accurate for a given computational complexity. In this paper, we show that the conventional particle swarm optimization approach, which was originally designed to solve general optimization problems where fast convergence of the algorithm might not be a primary concern, could be modified appropriately so that it could provide accurate motion estimation with very low computational cost in the specific context of video motion estimation. To this end, we proposed a new block matching algorithm based on a set of strategies adapted from the standard particle swarm optimization approach. Extensive simulations showed that the proposed method could achieve significant improvements over leading fast block matching methods including the diamond search and the cross-diamond search methods, in terms of both estimation accuracy and computational cost. In particular, the proposed method based on particle swarm optimization is not only much faster, but also remarkably more accurate (about 2dB higher in terms of the Peak Signal-to-Noise-Ratio) than the competing methods on video sequences with large motion.

The Insecurity of Wireless Networks

Wi-Fi is the standard protocol for wireless networks used extensively in US critical infrastructures. Since the Wired Equivalency Privacy (WEP) security protocol was broken, the Wi-Fi Protected Access (WPA) protocol has been considered the secure alternative compatible with hardware developed for WEP. However, in November 2008, researchers developed an attack on WPA, allowing forgery of Address Resolution Protocol (ARP) packets. Subsequent enhancements have enabled ARP poisoning, cryptosystem denial of service, and man-in-the-middle attacks. Open source systems and methods (OSSM) have long been used to secure networks against such attacks. This article reviews OSSMs and the results of experimental attacks on WPA. These experiments re-created current attacks in a laboratory setting, recording both wired and wireless traffic. The article discusses methods of intrusion detection and prevention in the context of cyberphysical protection of critical Internet infrastructure. The basis for this research is a specialized (and undoubtedly incomplete) taxonomy of Wi-Fi attacks and their adaptations to existing countermeasures and protocol revisions. Ultimately, this article aims to provide a clearer picture of how and why wireless protection protocols and encryption must achieve a more scientific basis for detecting and preventing such attacks.

Fast and accurate global motion estimation algorithm using pixel subsampling

Global motion generally describes the motion of a camera, although it may comprise motions of large objects. Global motions are often modeled by parametric transformations of two-dimensional images. The process of estimating the motions parameters is called global motion estimation (GME). GME is widely employed in many applications such as video coding, image stabilization and super-resolution. To estimate global motion parameters, the Levenburg-Marquardt algorithm (LMA) is typically used to minimize an objective function iteratively. Since the region of support for the global motion representation consists of the entire image frame, the minimization process tends to be very expensive computationally by involving all the pixels within an image frame. In order to significantly reduce the computational complexity of the LMA, we proposed to select only a small subset of the pixels for estimating the motion parameters, based on several subsampling patterns and their combinations. Simulation results demonstrated that the proposed method could speed up the conventional GME approach by over ten times, with only a very slight loss (less than 0.1dB) in estimation accuracy. The proposed method was also found to outperform several state-of-the-art fast GME methods in terms of the speed/accuracy tradeoffs.

Stationary queue-size distribution for variable complexity sequential decoders with large timeout

Fano decoders are convolutional channel decoders having variable decoding complexity in changing channel conditions. Buffers are usually required by the Fano decoder to store data blocks due to non-deterministic decoding delays. In practice, however, a predetermined limit is typically set for the decoding time so that the Fano decoder is not allowed to devote more time than the limit in decoding any single data block. We assume that data blocks arrive to the decoder with an infinite buffer according to a Bernoulli process, and that the decoding time of the Fano decoder follows the Pareto distribution parameterized by the signal-to-noise ratio (SNR) of the channel. If the channel has a smaller SNR, it typically takes longer time to decode a nosier data block. It is known in theory that if the SNR is lower than a certain threshold, the variance of the decoding time can go to infinity. In practice, if the decoding time of a block exceeds this limit, the block is removed from the buffer, resulting in a failure in decoding. In this paper, we present a method to obtain the stationary queue size distribution numerically for large decoding time limits.

Predictive Lossless Compression of Regions of Interest in Hyperspectral Images With No-Data Regions

This paper addresses the problem of efficient predictive lossless compression on the regions of interest (ROIs) in the hyperspectral images with no-data regions. We propose a two-stage prediction scheme, where a context-similarity-based weighted average prediction is followed by recursive least square filtering to decorrelate the hyperspectral images for compression. We then propose to apply separate Golomb-Rice codes for coding the prediction residuals of the full-context pixels and boundary pixels, respectively. To study the coding gains of this separate coding scheme, we introduce a mixture geometric model to represent the residuals associated with various combinations of the full-context pixels and boundary pixels. Both information-theoretic analysis and simulations on synthetic data confirm the advantage of the separate coding scheme over the conventional coding method based on a single underlying geometric distribution. We apply the aforementioned prediction and coding methods to four publicly available hyperspectral image data sets, attaining significant improvements over several other state-of-the-art methods, including the shape-adaptive JPEG 2000 method.

Efficient local transformation estimation using Lie operators

Conventional translation-only motion estimation algorithms cannot cope with transformations of objects such as scaling, rotations and deformations. Motion models characterizing non-translation motions are thus beneficial as they offer more accurate motion estimation and compensation. In this paper, we introduce low-complexity transformation estimation methods with four motion models based on Lie operators, which are linear operators that have found applications in optical character recognitions. We show that individual Lie operators are capable of capturing small degrees of object transformations. We propose an efficient local transformation estimation algorithm in order to further improve the accuracy of the translation-only estimation by integrating all four motion models. Simulations with an MPEG-2 video codec on two video sequences show that the proposed transformation estimation approach can noticeably improve the motion compensation performance of the translation-only method by achieving higher PSNR (peak signal-to-noise ratio) values for the predicted frames, with only a small fraction of the complexity required by the translation motion search.

A novel method for lossless compression of arbitrarily shaped regions of interest in hyperspectral imagery

We propose a novel algorithm for lossless compression of regions of interest (ROI) in hyperspectral images. The algorithm can compress arbitrarily shaped ROIs as specified by a binary map. The algorithm separates the boundary pixels from the full-context pixels within the ROI and applies Golomb-Rice encoders with different parameters on the boundary and full-context ROI pixels respectively. Experimental results show that the proposed algorithm provides larger compression than JPLs low-complexity hyperspectral image compressing method when applied on individual ROIs.

Five-step FFT algorithm with reduced computational complexity

We propose a fast Fourier transform algorithm, which removes two steps of twiddle factor multiplications from the conventional five-step FFT algorithm. The proposed FFT algorithm not only reduces the computational complexity of the five-step FFT algorithm by O(n) operations, but also reduces its memory requirement.

Evaluations of deep convolutional neural networks for automatic identification of malaria infected cells

This paper studied automatic identification of malaria infected cells using deep learning methods. We used whole slide images of thin blood stains to compile an dataset of malaria-infected red blood cells and non-infected cells, as labeled by a group of four pathologists. We evaluated three types of well-known convolutional neural networks, including the LeNet, AlexNet and GoogLeNet. Simulation results showed that all these deep convolution neural networks achieved classification accuracies of over 95%, higher than the accuracy of about 92% attainable by using the support vector machine method. Moreover, the deep learning methods have the advantage of being able to automatically learn the features from the input data, thereby requiring minimal inputs from human experts for automated malaria diagnosis.

Complexity accuracy tradeoffs of Lie operators in motion estimation

Lie operators are known to be effective in detecting very small degrees of object motions such as scaling, rotations and deformations. In this paper, we apply multiple Lie operators in combination to increase the accuracy of the conventional translation-only motion estimation. We propose the following three motion estimation methods using Lie operators with varying computational complexities, including the serial search, iterative search and the dynamic programming like search methods. We seek to study the tradeoffs allowed by the different combinations of Lie operators between the improved accuracy and the extra computational complexity associated with estimating the motion parameters. Both analytical and experimental results show that the proposed Lie-operator approaches can offer significant increases in the accuracy of the motion estimation with only low to moderate increase in the computational complexity. We also demonstrate that the iterative search method based on Lie operators has much lower complexity than motion search using an affine motion model for small motion parameters, while providing accuracy improvements very close to those attainable by the affine model approach.