Object detection is the first step in the object recognition. According to the detection results, its following works are affected. However, object detection has a heavy resource requirement in terms of, computing power and memory. If an image is enlarged, the computational load required for object detection is also increased. An-integral-image-based method guarantees fast object detection. Once an integral image is generated, the speed of the object detection procedure remains fixed, regardless of the pattern region size. However, this becomes an even greater issue if the image is enlarged. In this paper, we propose the use of directional integral image based object detection. A directional integral image gives direction to an integral image, which can then be calculated from various directions. Furthermore, many unnecessary calculations, which typically occur when a partial integral image is used for object detection, can be avoided. Therefore, the amount of computation is reduced, compared with methods using integral images. In experiments comparing methods, the proposed method required 40% fewer computations.

In this paper, we present several cryptanalyses of Hierocrypt-L1 block cipher, which was selected as one of the CRYPTREC recommended ciphers in Japan in 2003. We present a differential attack and an impossible differential attack on 8 S-function layers in a related-key setting. We first show that there exist the key scheduling differential characteristics which always hold, then we search for differential paths for the data randomizing part with the minimum active S-boxes using the above key differentials. We also show that our impossible differential attack is a new type.

In statistical methods, such as statistical static timing analysis, Gaussian mixture model (GMM) is a useful tool for representing a non-Gaussian distribution and handling correlation easily. In order to repeat various statistical operations such as summation and maximum for GMMs efficiently, the number of components should be restricted around two. In this paper, we propose a method for reducing the number of components of a given GMM to two (2-GMM). Moreover, since the distribution of each component is represented often by a linear combination of some explanatory variables, we propose a method to compute the covariance between each explanatory variable and the obtained 2-GMM, that is, the sensitivity of 2-GMM to each explanatory variable. In order to evaluate the performance of the proposed methods, we show some experimental results. The proposed methods minimize the normalized integral square error of probability density function of 2-GMM by the sacrifice of the accuracy of sensitivities of 2-GMM.

In this paper, we propose a novel decomposition method to segment multiple object regions simultaneously in cluttered videos. This method formulates object regions segmentation as a labeling problem in which we assign object IDs to the superpixels in a sequence of video frames so that the unary color matching cost is low, the assignment induces compact segments, and the superpixel labeling is consistent through time. Multi-object segmentation in a video is a combinatorial problem. We propose a binary linear formulation. Since the integer linear programming is hard to solve directly, we relax it and further decompose the relaxation into a sequence of much simpler max-flow problems. The proposed method is guaranteed to converge in a finite number of steps to the global optimum of the relaxation. It also has a high chance to obtain all integer solution and therefore achieves the global optimum. The rounding of the relaxation result gives an N-approximation solution, where N is the number of objects. Comparing to directly solving the integer program, the novel decomposition method speeds up the computation by orders of magnitude. Our experiments show that the proposed method is robust against object pose variation, occlusion and is more accurate than the competing methods while at the same time maintains the efficiency.

We propose subquadratic space complexity multipliers for any finite field $mathbb{F}_{q^n}$ over the base field $mathbb{F}_q$ using the Dickson basis, where q is a prime power. It is shown that a field multiplication in $mathbb{F}_{q^n}$ based on the Dickson basis results in computations of Toeplitz matrix vector products (TMVPs). Therefore, an efficient computation of a TMVP yields an efficient multiplier. In order to derive efficient $mathbb{F}_{q^n}$ multipliers, we develop computational schemes for a TMVP over $mathbb{F}_{q}$. As a result, the $mathbb{F}_{2^n}$ multipliers, as special cases of the proposed $mathbb{F}_{q^n}$ multipliers, have lower time complexities as well as space complexities compared with existing results. For example, in the case that n is a power of 3, the proposed $mathbb{F}_{2^n}$ multiplier for an irreducible Dickson trinomial has about 14% reduced space complexity and lower time complexity compared with the best known results.

We have proposed composite films composed of a molecular-aligned polymer and liquid crystal (LC) for substrate-free liquid crystal displays with high-contrast images. We successfully controlled the molecular alignment of the LC and formed molecular-aligned LC droplets in the polymer by controlling the fluidity of the LC/monomer mixture and the curing rate of the monomer.

This paper proposes an efficient lossless coding scheme for color video in RGB 4:4:4 format. For the R signal that is encoded before the other signals at each frame, we employ a block-adaptive prediction technique originally developed for monochrome video. The prediction technique used for the remaining G and B signals is extended to exploit inter-color correlations as well as inter- and intra-frame ones. In both cases, multiple predictors are adaptively selected on a block-by-block basis. For the purpose of designing a set of predictors well suited to the local properties of video signals, we also explore an appropriate setting for the spatiotemporal partitioning of a video volume.

We propose a speech analysis method based on the source-filter model using multivariate empirical mode decomposition (MEMD). The proposed method takes multiple adjacent frames of a speech signal into account by combining their log spectra into multivariate signals. The multivariate signals are then decomposed into intrinsic mode functions (IMFs). The IMFs are divided into two groups using the peak of the autocorrelation function (ACF) of an IMF. The first group characterized by a spectral fine structure is used to estimate the fundamental frequency F0 by using the ACF, whereas the second group characterized by the frequency response of the vocal-tract filter is used to estimate formant frequencies by using a peak picking technique. There are two advantages of using MEMD: (i) the variation in the number of IMFs is eliminated in contrast with single-frame based empirical mode decomposition and (ii) the common information of the adjacent frames aligns in the same order of IMFs because of the common mode alignment property of MEMD. These advantages make the analysis more accurate than with other methods. As opposed to the conventional linear prediction (LP) and cepstrum methods, which rely on the LP order and cut-off frequency, respectively, the proposed method automatically separates the glottal-source and vocal-tract filter. The results showed that the proposed method exhibits the highest accuracy of F0 estimation and correctly estimates the formant frequencies of the vocal-tract filter.

A novel method for illumination-invariant face representation is presented based on the orthogonal decomposition of the local image structure. One important advantage of the proposed method is that image gradients and corresponding intensity values are simultaneously used with our decomposition procedure to preserve the original texture while yielding the illumination-invariant feature space. Experimental results demonstrate that the proposed method is effective for face recognition and verification even with diverse lighting conditions.

The introduction of deep neural networks (DNNs) leads to a significant improvement of the automatic speech recognition (ASR) performance. However, the whole ASR system remains sophisticated due to the dependent on the hidden Markov model (HMM). Recently, a new end-to-end ASR framework, which utilizes recurrent neural networks (RNNs) to directly model context-independent targets with connectionist temporal classification (CTC) objective function, is proposed and achieves comparable results with the hybrid HMM/DNN system. In this paper, we investigate per-dimensional learning rate methods, ADAGRAD and ADADELTA included, to improve the recognition of the end-to-end system, based on the fact that the blank symbol used in CTC technique dominates the output and these methods give frequent features small learning rates. Experiment results show that more than 4% relative reduction of word error rate (WER) as well as 5% absolute improvement of label accuracy on the training set are achieved when using ADADELTA, and fewer epochs of training are needed.

In this letter, a novel channel impulse response (CIR)-based fingerprinting positioning method using kernel principal component analysis (KPCA) has been proposed. During the offline phase of the proposed method, a survey is performed to collect all CIRs from access points, and a fingerprint database is constructed, which has vectors including CIR and physical location. During the online phase, KPCA is first employed to solve the nonlinearity and complexity in the CIR-position dependencies and extract the principal nonlinear features in CIRs, and support vector regression is then used to adaptively learn the regress function between the KPCA components and physical locations. In addition, the iterative narrowing-scope step is further used to refine the estimation. The performance comparison shows that the proposed method outperforms the traditional received signal strength based positioning methods.

We propose an edge-preserving multiscale image decomposition method using filters for non-equispaced signals. It is inspired by the domain transform, which is a high-speed edge-preserving smoothing method, and it can be used in many image processing applications. One of the disadvantages of the domain transform is sensitivity to noise. Even though the proposed method is based on non-equispaced filters similar to the domain transform, it is robust to noise since it employs a multiscale decomposition. It uses the Laplacian pyramid scheme to decompose an input signal into the piecewise-smooth components and detail components. We design the filters by using an optimization based on edge-preserving smoothing with a conversion of signal distances and filters taking into account the distances between signal intervals. In addition, we also propose construction methods of filters for non-equispaced signals by using arbitrary continuous filters or graph spectral filters in order that various filters can be accommodated by the proposed method. As expected, we find that, similar to state-of-the-art edge-preserving smoothing techniques, including the domain transform, our approach can be used in many applications. We evaluated its effectiveness in edge-preserving smoothing of noise-free and noisy images, detail enhancement, pencil drawing, and stylization.

This paper proposes a blind, inaudible, robust digital-audio watermarking scheme based on singular-spectrum analysis, which relates to watermarking techniques based on singular value decomposition. We decompose a host signal into its oscillatory components and modify amplitudes of some of those components with respect to a watermark bit and embedding rule. To improve the sound quality of a watermarked signal and still maintain robustness, differential evolution is introduced to find optimal parameters of the proposed scheme. Test results show that, although a trade-off between inaudibility and robustness still persists, the difference in sound quality between the original and the watermarked one is considerably smaller. This improved scheme is robust against many attacks, such as MP3 and MP4 compression, and band-pass filtering. However, there is a drawback, i.e., some music-dependent parameters need to be shared between embedding and extraction processes. To overcome this drawback, we propose a method for automatic parameter estimation. By incorporating the estimation method into the framework, those parameters need not to be shared, and the test results show that it can blindly decode watermark bits with an accuracy of 99.99%. This paper not only proposes a new technique and scheme but also discusses the singular value and its physical interpretation.

Estimation of the time delay of arrival (TDOA) problem is important to acoustic source localization. The TDOA estimation problem is defined as finding the relative delay between several microphone signals for the direct sound. To estimate TDOA, the generalized cross-correlation (GCC) method is the most frequently used, but it has a disadvantage in terms of reverberant environments. In order to overcome this problem, the adaptive eigenvalue decomposition (AED) method has been developed, which estimates the room transfer function and finds the direct-path delay. However, the algorithm does not take into account the fact that the room transfer function is a sparse channel, and so sometimes the estimated transfer function is too dense, resulting in failure to exact direct-path and delay. In this paper, an enhanced AED algorithm that makes use of a proportionate step-size control and a direct-path constraint is proposed instead of a constant step size and the L2-norm constraint. The simulation results show that the proposed algorithm has enhanced performance as compared to both the conventional AED method and the phase-transform (PHAT) algorithm.

Energy-efficient resource allocation is considered in sensing-based spectrum sharing for cooperative cognitive radio networks (CCRNs). The secondary user first listens to the spectrum allocated to the primary user (PU) to detect the PU state and then initiates data transmission with two power levels based on the sensing decision (e.g., idle or busy). Under this model, the optimization problem of maximizing energy efficiency (EE) is formulated over the transmission power and sensing time subject to some practical limitations, such as the individual power constraint for secondary source and relay, the quality of service (QoS) for the secondary system, and effective protection for the PU. Given the complexity of this problem, two simplified versions (i.e., perfect and imperfect sensing cases) are studied in this paper. We transform the considered problem in fractional form into an equivalent optimization problem in subtractive form. Then, for perfect sensing, the Lagrange dual decomposition and iterative algorithm are applied to acquire the optimal power allocation policy; for imperfect sensing, an exhaustive search and iterative algorithm are proposed to obtain the optimal sensing time and corresponding power allocation strategy. Finally, numerical results show that the energy-efficient design greatly improves EE compared with the conventional spectrum-efficient design.

In this paper, we propose a single-channel speech enhancement method for a push-to-talk enabled wireless communication device. The proposed method is based on adaptive weighted β-order spectral amplitude estimation under speech presence uncertainty and enhanced instantaneous phase estimation in order to achieve flexible and effective noise reduction while limiting the speech distortion due to different noise conditions. Experimental results confirm that the proposed method delivers higher voice quality and intelligibility than the reference methods in various noise environments.

An efficient Monte Carlo (MC) method for the calculation of failure probability degradation of an SRAM cell due to negative bias temperature instability (NBTI) is proposed. In the proposed method, a particle filter is utilized to incrementally track temporal performance changes in an SRAM cell. The number of simulations required to obtain stable particle distribution is greatly reduced, by reusing the final distribution of the particles in the last time step as the initial distribution. Combining with the use of a binary classifier, with which an MC sample is quickly judged whether it causes a malfunction of the cell or not, the total number of simulations to capture the temporal change of failure probability is significantly reduced. The proposed method achieves 13.4× speed-up over the state-of-the-art method.

Most unsupervised video segmentation algorithms are difficult to handle object extraction in dynamic real-world scenes with large displacements, as foreground hypothesis is often initialized with no explicit mutual constraint on top-down spatio-temporal coherency despite that it may be imposed to the segmentation objective. To handle such situations, we propose a multiscale saliency flow (MSF) model that jointly learns both foreground and background features of multiscale salient evidences, hence allowing temporally coherent top-down information in one frame to be propagated throughout the remaining frames. In particular, the top-down evidences are detected by combining saliency signature within a certain range of higher scales of approximation coefficients in wavelet domain. Saliency flow is then estimated by Gaussian kernel correlation of non-maximal suppressed multiscale evidences, which are characterized by HOG descriptors in a high-dimensional feature space. We build the proposed MSF model in accordance with the primary object hypothesis that jointly integrates temporal consistent constraints of saliency map estimated at multiple scales into the objective. We demonstrate the effectiveness of the proposed multiscale saliency flow for segmenting dynamic real-world scenes with large displacements caused by uniform sampling of video sequences.

The separation of signals with temporal structure from mixed sources is a challenging problem in signal processing. For this problem, blind source extraction (BSE) is more suitable than blind source separation (BSS) because it has lower computation cost. Nowadays many BSE algorithms can be used to extract signals with temporal structure. However, some of them are not robust because they are too dependent on the estimation precision of time delay; some others need to choose parameters before extracting, which means that arbitrariness can't be avoided. In order to solve the above problems, we propose a robust source extraction algorithm whose performance doesn't rely on the choice of parameters. The algorithm is realized by maximizing the objective function that we develop based on the non-Gaussianity and the temporal structure of source signals. Furthermore, we analyze the stability of the algorithm. Simulation results show that the algorithm can extract the desired signal from large numbers of observed sensor signals and is very robust to error in the estimation of time delay.

As technology nodes continue to shrink, the impact of radiation-induced soft error on processor reliability increases. Estimation of processor reliability and identification of vulnerable flip-flops requires accurate soft error rate (SER) analysis techniques. This paper presents a proposal for a soft error propagation analysis technique. We specifically examine single event upset (SEU) occurring at a flip-flop in sequential circuits. When SEUs propagate in sequential circuits, the faults can be masked temporally and logically. Conventional soft error propagation analysis techniques do not consider error convergent timing on re-convergent paths. The proposed technique can analyze soft error propagation while considering error-convergent timing on a re-convergent path by combinational analysis of temporal and logical effects. The proposed technique also considers the case in which the temporal masking is disabled with an enable signal of the erroneous flip-flop negated. Experimental results show that the proposed technique improves inaccuracy by 70.5%, on average, compared with conventional techniques using ITC 99 and ISCAS 89 benchmark circuits when the enable probability is 1/3, while the runtime overhead is only 1.7% on average.