Lianqiang LI Jie ZHU Ming-Ting SUN
Convolutional Neural Networks (CNNs) usually have millions or even billions of parameters, which make them hard to be deployed into mobile devices. In this work, we present a novel filter-level pruning method to alleviate this issue. More concretely, we first construct an undirected fully connected graph to represent a pre-trained CNN model. Then, we employ the spectral clustering algorithm to divide the graph into some subgraphs, which is equivalent to clustering the similar filters of the CNN into the same groups. After gaining the grouping relationships among the filters, we finally keep one filter for one group and retrain the pruned model. Compared with previous pruning methods that identify the redundant filters by heuristic ways, the proposed method can select the pruning candidates more reasonably and precisely. Experimental results also show that our proposed pruning method has significant improvements over the state-of-the-arts.
A non-photorealistic rendering method has been proposed for generating oil-film-like images from photographic images by bilateral infra-envelope filter. The conventional method has a disadvantage that it takes much time to process. We propose a method for generating oil-film-like images that can be processed faster than the conventional method. The proposed method uses an iterative process with upper and lower smoothing filters. To verify the effectiveness of the proposed method, we conduct experiments using Lenna image. As a result of the experiments, we show that the proposed method can process faster than the conventional method.
Kiyoshi NISHIYAMA Masahiro SUNOHARA Nobuhiko HIRUMA
The least mean squares (LMS) algorithm has been widely used for adaptive filtering because of easily implementing at a computational complexity of O(2N) where N is the number of taps. The drawback of the LMS algorithm is that its performance is sensitive to the scaling of the input. The normalized LMS (NLMS) algorithm solves this problem on the LMS algorithm by normalizing with the sliding-window power of the input; however, this normalization increases the computational cost to O(3N) per iteration. In this work, we derive a new formula to strictly perform the NLMS algorithm at a computational complexity of O(2N), that is referred to as the C-NLMS algorithm. The derivation of the C-NLMS algorithm uses the H∞ framework presented previously by one of the authors for creating a unified view of adaptive filtering algorithms. The validity of the C-NLMS algorithm is verified using simulations.
Akira KUBOTA Kazuya KODAMA Asami ITO
A pupil function of aperture in image capturing systems is theoretically derived such that one can perfectly reconstruct all-in-focus image through linear filtering of the focal stack. The perfect reconstruction filters are also designed based on the derived pupil function. The designed filters are space-invariant; hence the presented method does not require region segmentation. Simulation results using synthetic scenes shows effectiveness of the derived pupil function and the filters.
Zuopeng ZHAO Hongda ZHANG Yi LIU Nana ZHOU Han ZHENG Shanyi SUN Xiaoman LI Sili XIA
Although correlation filter-based trackers have demonstrated excellent performance for visual object tracking, there remain several challenges to be addressed. In this work, we propose a novel tracker based on the correlation filter framework. Traditional trackers face difficulty in accurately adapting to changes in the scale of the target when the target moves quickly. To address this, we suggest a scale adaptive scheme based on prediction scales. We also incorporate a speed-based adaptive model update method to further improve overall tracking performance. Experiments with samples from the OTB100 and KITTI datasets demonstrate that our method outperforms existing state-of-the-art tracking algorithms in fast motion scenes.
This paper deals with the problem of minimizing roundoff noise and pole sensitivity simultaneously subject to l2-scaling constraints for state-space digital filters. A novel measure for evaluating roundoff noise and pole sensitivity is proposed, and an efficient technique for minimizing this measure by jointly optimizing state-space realization and error feedback is explored, namely, the constrained optimization problem at hand is converted into an unconstrained problem and then the resultant problem is solved by employing a quasi-Newton algorithm. A numerical example is presented to demonstrate the validity and effectiveness of the proposed technique.
Peng CHEN Weijun LI Linjun SUN Xin NING Lina YU Liping ZHANG
Human gender recognition in the wild is a challenging task due to complex face variations, such as poses, lighting, occlusions, etc. In this letter, learnable Gabor convolutional network (LGCN), a new neural network computing framework for gender recognition was proposed. In LGCN, a learnable Gabor filter (LGF) is introduced and combined with the convolutional neural network (CNN). Specifically, the proposed framework is constructed by replacing some first layer convolutional kernels of a standard CNN with LGFs. Here, LGFs learn intrinsic parameters by using standard back propagation method, so that the values of those parameters are no longer fixed by experience as traditional methods, but can be modified by self-learning automatically. In addition, the performance of LGCN in gender recognition is further improved by applying a proposed feature combination strategy. The experimental results demonstrate that, compared to the standard CNNs with identical network architecture, our approach achieves better performance on three challenging public datasets without introducing any sacrifice in parameter size.
Zhangkai LUO Zhongmin PEI Bo ZOU
In this letter, a polarization filtering based transmission (PFBT) scheme is proposed to enhance the spectrum efficiency in wireless communications. In such scheme, the information is divided into several parts and each is conveyed by a polarized signal with a unique polarization state (PS). Then, the polarized signals are added up and transmitted by the dual-polarized antenna. At the receiver side, the oblique projection polarization filters (OPPFs) are adopted to separate each polarized signal. Thus, they can be demodulated separately. We mainly focus on the construction methods of the OPPF matrix when the number of the separate parts is 2 and 3 and evaluate the performance in terms of the capacity and the bit error rate. In addition, we also discuss the probability of the signal separation when the number of the separate parts is equal or greater than 4. Theoretical results and simulation results demonstrate the performance of the proposed scheme.
Xiang LU Ziyang CHEN Lianpo WANG Ruidong LI Chao ZHAI
In resent years, providing location services for mobile targets in a closed environment has been a growing interest. In order to provide good localization and tracking performance for drones in GPS-denied scenarios, this paper proposes a multi-tag radio frequency identification (RFID) system that is easy to equip and does not take up the limited resources of the drone which is not susceptible to processor performance and cost constraints compared with computer vision based approaches. The passive RFID tags, no battery equipped, have an ultra-high resolution of millimeter level. We attach multiple tags to the drone and form multiple sets of virtual antenna arrays during motion, avoiding arranging redundant antennas in applications, and calibrating the speed chain to improve tracking performance. After combining the strap-down inertial navigation system (SINS) carried by the drone, we have established a coupled integration model that can suppress the drift error of SINS with time. The experiment was designed in bi-dimensional and three-dimensional scenarios, and the integrated positioning system based on SINS/RFID was evaluated. Finally, we discussed the impact of some parameters, this innovative approach is verified in real scenarios.
Chengcheng JIANG Xinyu ZHU Chao LI Gengsheng CHEN
Pre-trained CNNs on ImageNet have been widely used in object tracking for feature extraction. However, due to the domain mismatch between image classification and object tracking, the submergence of the target-specific features by noise largely decreases the expression ability of the convolutional features, resulting in an inefficient tracking. In this paper, we propose a robust tracking algorithm with low-dimensional target-specific feature extraction. First, a novel cascaded PCA module is proposed to have an explicit extraction of the low-dimensional target-specific features, which makes the new appearance model more effective and efficient. Next, a fast particle filter process is raised to further accelerate the whole tracking pipeline by sharing convolutional computation with a ROI-Align layer. Moreover, a classification-score guided scheme is used to update the appearance model for adapting to target variations while at the same time avoiding the model drift that caused by the object occlusion. Experimental results on OTB100 and Temple Color128 show that, the proposed algorithm has achieved a superior performance among real-time trackers. Besides, our algorithm is competitive with the state-of-the-art trackers in precision while runs at a real-time speed.
In this paper, a reduction of the number of components included in direct simulation type active complex filter is proposed. The proposed method is achieved by sharing NIC's (Negative Impedance Converters) which satisfy some conditions. Compared with the conventional method, the proposed one has wide generality. As an example, a third-order complex elliptic filter is designed. The validity of the proposed method is confirmed through experiment.
Yuanlei QI Feiran YANG Ming WU Jun YANG
The blind multichannel identification is useful in many applications. Although many approaches have been proposed to address this challenging problem, the adaptive filtering-based methods are attractive due to their computational efficiency and good convergence property. The multichannel normalized least mean-square (MCNLMS) algorithm is easy to implement, but it converges very slowly for a correlated input. The multichannel affine projection algorithm (MCAPA) is thus proposed to speed up the convergence. However, the convergence of the MCNLMS and MCAPA is still unsatisfactory in practice. In this paper, we propose a time-domain Kalman filtering approach to the blind multichannel identification problem. Specifically, the proposed adaptive Kalman filter is based on the cross relation method and also uses more past input vectors to explore the decorrelation property. Simulation results indicate that the proposed method outperforms the MCNLMS and MCAPA significantly in terms of the initial convergence and tracking capability.
Haosheng ZHANG Aravind THARAYIL NARAYANAN Hans HERDIAN Bangan LIU Rui WU Atsushi SHIRANE Kenichi OKADA
This paper presents a high power efficient pulse VCO with tail-filter for the chip-scale atomic clock (CSAC) application. The stringent power and clock stability specifications of next-generation CSAC demand a VCO with ultra-low power consumption and low phase noise. The proposed VCO architecture aims for the high power efficiency, while further reducing the phase noise using tail filtering technique. The VCO has been implemented in a standard 45nm SOI technology for validation. At an oscillation frequency of 5.0GHz, the proposed VCO achieves a phase noise of -120dBc/Hz at 1MHz offset, while consuming 1.35mW. This translates into an FoM of -191dBc/Hz.
Taiki SHINOHARA Takashi YOSHIDA Naoyuki AIKAWA
Two-dimensional (2-D) maximally flat finite impulse response (FIR) digital filters have flat characteristics in both passband and stopband. 2-D maximally flat diamond-shaped half-band FIR digital filter can be designed very efficiently as a special case of 2-D half-band FIR filters. In some cases, this filter would require the reduction of the filter lengths for one of the axes while keeping the other axis unchanged. However, the conventional methods can realize such filters only if difference between each order is 2, 4 and 6. In this paper, we propose a closed-form frequency response of 2-D low-pass maximally flat diamond-shaped half-band FIR digital filters with arbitrary filter orders. The constraints to treat arbitrary filter orders are firstly proposed. Then, a closed-form transfer function is achieved by using Bernstein polynomial.
This paper presents a novel delta-sigma modulator that uses a switched-capacitor (SC) integrator with the structure of a finite impulse response (FIR) filter in a loop filter configuration. The delta-sigma analog-to-digital converter (ΔΣADC) is used in various conversion systems to enable low-power, high-accuracy conversion using oversampling and noise shaping. Increasing the gain coefficient of the integrator in the loop filter configuration of the ΔΣADC suppresses the quantization noise that occurs in the signal band. However, there is a trade-off relationship between the integrator gain coefficient and system stability. The SC integrator, which contains an FIR filter, can suppress quantization noise in the signal band without requiring an additional operational amplifier. Additionally, it can realize a higher signal-to-quantization noise ratio. In addition, the poles that are added by the FIR filter structure can improve the system's stability. It is also possible to improve the flexibility of the pole placement in the system. Therefore, a noise transfer function that does not contain a large gain peak is realized. This results in a stable system operation. This paper presents the essential design aspects of a ΔΣADC with an FIR filter. Two types of simulation results are examined for the proposed first- and second-order, and these results confirm the effectiveness of the proposed architecture.
Song BIAN Masayuki HIROMOTO Takashi SATO
In this work, we provide the first practical secure email filtering scheme based on homomorphic encryption. Specifically, we construct a secure naïve Bayesian filter (SNBF) using the Paillier scheme, a partially homomorphic encryption (PHE) scheme. We first show that SNBF can be implemented with only the additive homomorphism, thus eliminating the need to employ expensive fully homomorphic schemes. In addition, the design space for specialized hardware architecture realizing SNBF is explored. We utilize a recursive Karatsuba Montgomery structure to accelerate the homomorphic operations, where multiplication of 2048-bit integers are carried out. Through the experiment, both software and hardware versions of the SNBF are implemented. On software, 104-105x runtime and 103x storage reduction are achieved by SNBF, when compared to existing fully homomorphic approaches. By instantiating the designed hardware for SNBF, a further 33x runtime and 1919x power reduction are achieved. The proposed hardware implementation classifies an average-length email in under 0.5s, which is much more practical than existing solutions.
Hiroshi SEKI Kazumasa YAMAMOTO Tomoyosi AKIBA Seiichi NAKAGAWA
Deep neural networks (DNNs) have achieved significant success in the field of automatic speech recognition. One main advantage of DNNs is automatic feature extraction without human intervention. However, adaptation under limited available data remains a major challenge for DNN-based systems because of their enormous free parameters. In this paper, we propose a filterbank-incorporated DNN that incorporates a filterbank layer that presents the filter shape/center frequency and a DNN-based acoustic model. The filterbank layer and the following networks of the proposed model are trained jointly by exploiting the advantages of the hierarchical feature extraction, while most systems use pre-defined mel-scale filterbank features as input acoustic features to DNNs. Filters in the filterbank layer are parameterized to represent speaker characteristics while minimizing a number of parameters. The optimization of one type of parameters corresponds to the Vocal Tract Length Normalization (VTLN), and another type corresponds to feature-space Maximum Linear Likelihood Regression (fMLLR) and feature-space Discriminative Linear Regression (fDLR). Since the filterbank layer consists of just a few parameters, it is advantageous in adaptation under limited available data. In the experiment, filterbank-incorporated DNNs showed effectiveness in speaker/gender adaptations under limited adaptation data. Experimental results on CSJ task demonstrate that the adaptation of proposed model showed 5.8% word error reduction ratio with 10 utterances against the un-adapted model.
Shogo KOYANAGI Teruyuki MIYAJIMA
In this paper, we consider full-duplex (FD) relay networks with filter-and-forward (FF)-based multiple relays (FD-FF), where relay filters jointly mitigate self-interference (SI), inter-relay interference (IRI), and inter-symbol interference. We consider the filter design problem based on signal-to-noise-plus-interference ratio maximization subject to a total relay transmit power constraint. To make the problem tractable, we propose two methods: one that imposes an additional constraint whereby the filter responses to SI and IRI are nulled, and the other that makes i.i.d. assumptions on the relay transmit signals. Simulation results show that the proposed FD-FF scheme outperforms a conventional FF scheme in half-duplex mode. We also consider the filter design when only second-order statistics of channel path gains are available.
Koichi ICHIGE Nobuya ARAKAWA Ryo SAITO Osamu SHIBATA
This paper presents a radio-based real-time moving object tracking method based on Kalman filtering using a phase-difference compensation technique and a non-uniform pulse transmission scheme. Conventional Kalman-based tracking methods often require time, amplitude, phase information and their derivatives for each receiver antenna; however, their location estimation accuracy does not become good even with many transmitting pulses. The presented method employs relative phase-difference information and a non-uniform pulse generation scheme, which can greatly reduce the number of transmitting pulses while preserving the tracking accuracy. Its performance is evaluated in comparison with that of conventional methods.
Zi Hao ONG Takahide SATO Satomi OGAWA
A design method of the differential N-path filter with sampling computation is proposed. It enables the scale of the whole filter to be reduced by approximately half for easier realization. On top of that, the proposed method offers the ability to eliminate the harmonic passbands of the clock frequency and an increase of harmonic rejection. By using the proposed method, previous work involving an 8-path filter can be reduced to 5-path. The proposed differential 5-path filter reduces the scale of the circuit and at the same time has the performance of a 10-path filter from previous work. An example of differential 7-path filter using the same proposed design method is also stated in comparison of the differential 5-path filter. The differential 7-path filter offers the ability to eliminate all the passbands below 10 times the clock frequency with a tradeoff of an increase in circuit scale.