Neighborhood Preserving Embedding (NPE) is a linear approximate of Locally Linear Embedding with local geometry preserving property. An efficient face recognition method requires a mapping to separate within-class structure from between-class structure. But, NPE is not in the case since it is unsupervised. Hence, we improved NPE- utilizing the neighbor and class relations of face data. The proposed technique, Neighborhood Discriminant Embedding (NDE), takes into account the local neighboring and the between-class neighboring information. NDE seeks a projection in such that the projected samples of different classes are far apart. Experimental results show that NDE achieves superior results to NPE in face authentication with error reduction at least 20%.
A chaotic integrated circuit is designed and fabricated using a 0.35µm CMOS process. The circuit iterates an N-shaped transfer function using a small analog neural network. One of the advantages of the proposed circuit is its small circuit area with only 13MOS transistors. The circuit generates both an analog and a digital signal that can be used to create true random bit sequences.
An area and power efficient signal reordering unit (SRU) for OFDM systems that reorders the bit-reversed output of an FFT unit into linear order is proposed. By using a FIFO that is smaller than an N-word memory used in conventional SRUs, the proposed SRU achieves about 25% reductions in both the area and the power consumption in case of a 1024-point FFT.
This paper presents a sound source localization algorithm based on zero crossings in acoustic reverberant environments. From sudden increases of acoustic energy in a non-stationary signal, we identify intervals which dominantly include signal components through direct paths from a source to sensors because they may provide reliable localization cues corresponding to the source direction despite the reverberation. Experimental results show that the presented algorithm which is based on zero crossings with the detection of direct components efficiently accomplishes sound source localization in reverberant environments.
Face recognition system usually consists of feature extraction and pattern classification. However, not all of extracted facial features contribute to the classification positively because of the variations of illumination and poses in face images. In this paper, an evolutionary feature selection algorithm is proposed in which discrete cosine transform (DCT) and genetic algorithms (GAs) are utilized to create a framework of feature acquisition. In detail, the face images are first transformed to frequency domain through DCT, then GAs are used to seek for optimal features in the redundant DCT coefficients where the generalization performance guides the searching process. Furthermore, an entropy-based extension on proposed evolving feature selection method is presented. In experiments, two face databases are used to evaluate the effectiveness of our proposals.
In this study, a new, simple and accurate computation of the received signal strength (RSS) level for indoor environment is performed. The genetic algorithm (GA) approach is used for prediction of the RSS. The proposed model is formed on the knowledge of measurements without requiring any detail of the environment. The model provides a time efficient method to estimate RSS dynamically at any location in the test environment. The accuracy of the measurement results and the genetic algorithm approach are presented for three distinct transmitters located at different positions.
A theoretical prediction of beyond-THz frequency operation is demonstrated for a III-nitride heterojunction FET with a gate length of sub-100nm. The calculation is based upon an ensemble Monte Carlo simulation coupled with a 2D Poisson equation. The simulation results suggest that a current gain cutoff frequency of more than 1THz is achieved by an AlInN/InN/AlInN or AlInN/InGaN/AlInN double-heterojunction FET with a gate length of less than 50nm or 30nm, respectively, which are fabricated on a non-polar GaN substrate. The importance of high-mobility InN and InGaN used as a channel material for high-speed and high-frequency applications is numerically verified.
In this paper, we propose an object recognition technique based on particle swarm optimization. Our technique reduces the complexity of the Hough transform and the rigidity of template matching via cross correlation, while harnessing the robustness of both methods. To demonstrate its effectiveness, the proposed technique has been applied for ellipsoid detection in three-dimensional images and satisfactory results have been obtained. The detected object is described by a set of parameters, which can be used for further processing. In addition, our technique can be adapted for the detection of other objects, provided that they can be represented by a set of parameters.
A novel third order delta sigma modulator (DSM) is presented. The third order loop filter in the proposed DSM shares one opamp among three integrator stages through an optimal operation timing, which makes use of load capacitance differences between integrator stages. The power dissipation of the proposed DSM is much lower than conventional third order DSM because power consuming blocks in the DSM are only one opamp and a comparator. The proposed DSM, designed on 0.35µm CMOS process under 3.3V supply achieves 1mW power dissipation with 100-11kHz band width and 96dB SQNR.