Showing papers by "Jeng-Shyang Pan published in 2017"

Superimposed Sparse Parameter Classifiers for Face Recognition

Abstract: In this paper, a novel classifier, called superimposed sparse parameter (SSP) classifier is proposed for face recognition. SSP is motivated by two phase test sample sparse representation (TPTSSR) and linear regression classification (LRC), which can be treated as the extended of sparse representation classification (SRC). SRC uses all the train samples to produce the sparse representation vector for classification. The LRC, which can be interpreted as L2-norm sparse representation, uses the distances between the test sample and the class subspaces for classification. TPTSSR is also L2-norm sparse representation and uses two phase to compute the distance for classification. Instead of the distances, the SSP classifier employs the SSPs, which can be expressed as the sum of the linear regression parameters of each class in iterations, is used for face classification. Further, the fast SSP (FSSP) classifier is also suggested to reduce the computation cost. A mass of experiments on Georgia Tech face database, ORL face database, CVL face database, AR face database, and CASIA face database are used to evaluate the proposed algorithms. The experimental results demonstrate that the proposed methods achieve better recognition rate than the LRC, SRC, collaborative representation-based classification, regularized robust coding, relaxed collaborative representation, support vector machine, and TPTSSR for face recognition under various conditions.

Monkey King Evolution: an enhanced ebb-tide-fish algorithm for global optimization and its application in vehicle navigation under wireless sensor network environment

Abstract: Optimization algorithms are proposed to maximize the desirable properties while simultaneously minimizing the undesirable characteristics. Particle Swarm Optimization (PSO) is a famous optimization algorithm, and it has undergone many variants since its inception in 1995. Though different topologies and relations among particles are used in some state-of-the-art PSO variants, the overall performance on high dimensional multimodal optimization problem is still not very good. In this paper, we present a new memetic optimization algorithm, named Monkey King Evolutionary (MKE) algorithm, and give a comparative view of the PSO variants, including the canonical PSO, Inertia Weighted PSO, Constriction Coefficients PSO, Fully-Informed Particle Sawrm, Cooperative PSO, Comprehensive Learning PSO and some variants proposed in recent years, such as Dynamic Neighborhood Learning PSO, Social Learning Particle Swarm Optimization etc. The proposed MKE algorithm is a further work of ebb-tide-fish algorithm and what's more it performs very well not only on unimodal benchmark functions but also on multimodal ones on high dimensions. Comparison results under CEC2013 test suite for real parameter optimization show that the proposed MKE algorithm outperforms state-of-the-art PSO variants significantly. An application of the vehicle navigation optimization is also discussed in the paper, and the conducted experiment shows that the proposed approach to path navigation optimization saves travel time of real-time traffic navigation in a micro-scope traffic networks.

A Provable Secure Private Data Delegation Scheme for Mountaineering Events in Emergency System

Abstract: Recently, the sport of mountaineering is a popular leisure activity and many people may injure while mountaineering. In the year of 2014, Chen et al. suggested a cloud-based emergency response and SOS system for mountaineering travelers when they encounter dangers. Chen et al. claimed that their proposed system is secure against various known attacks and the executive performance of the system is reasonable when the protocol is implemented on the traveler’s mobile device. However, in this paper, we discover that Chen et al. ’s scheme is unable to protect the privacy of mountaineering travelers and the vulnerability allows a malicious attacker to spy on the electronic medical records of all mountaineering travelers by launching eavesdropping attacks. Moreover, Chen et al. ’s scheme is vulnerable to off-line password guessing attack when the mobile device of the mountaineering traveler is lost or stolen by an attacker. In order to repair these shortcomings existing in Chen et al. ’s scheme, we suggest an improved version of their scheme, which is provably secure in the random oracle model under the DDH and CDH problems.

A segment-based approach for large-scale ontology matching

Abstract: The most ground approach to solve the ontology heterogeneous problem is to determine the semantically identical entities between them, so-called ontology matching. However, the correct and complete identification of semantic correspondences is difficult to achieve with the scale of the ontologies that are huge; thus, achieving good efficiency is the major challenge for large- scale ontology matching tasks. On the basis of our former work, in this paper, we further propose a scalable segment-based ontology matching framework to improve the efficiency of matching large-scale ontologies. In particular, our proposal first divides the source ontology into several disjoint segments through an ontology partition algorithm; each obtained source segment is then used to divide the target ontology by a concept relevance measure; finally, these similar ontology segments are matched in a time and aggregated into the final ontology alignment through a hybrid Evolutionary Algorithm. In the experiment, testing cases with different scales are used to test the performance of our proposal, and the comparison with the participants in OAEI 2014 shows the effectiveness of our approach.

Clustering Formation in Wireless Sensor Networks: A Survey.

Abstract: Clustering formation is the modern energy efficient techniques in the designing and implementing Wireless sensor networks (WSN). Clustering provides various advantages like energy efficiency, lifetime, scalability, and less delay; but it leads to hot spot problem. The proposed unequal clustering is to overcome this issue. In uneven clustering, the cluster size varies proportionally to the distance to the base station (BS). In this paper, a comprehensive survey of various clustering formation approaches with their objectives, characteristics, etc. is presented. Also, the classifications of uneven clustering methods are carried out and compared them based on various cluster properties, Cluster Head (CH) properties, and clustering process.

A genetic-based effective approach to path-planning of autonomous underwater glider with upstream-current avoidance in variable oceans

Abstract: In this work, an exponential effective function (EEF) is developed as fitness function applied in a hybrid-Genetic Algorithm (hybrid-GA) to propose a genetic-based effective approach to the glider path-planning of ocean-sampling mission in variable oceans. The proposed EEF is such an objective function that is able to be implemented in optimization algorithm such as Genetic Algorithm (GA) for evaluation of the fittest path. In consideration of the glider path-planning problem (GPP), two motivations are driven by the proposed approach to the glider path-planning in discovery of: (1) a reachable path with the upstream-current avoidance (UCA) in variable oceans; (2) an efficient path for the glider ocean-sampling mission. The exponential combination of the glider motion and current effects as well as the cruising distance benefits the path in terms of reachability and efficiency. The reachability is the first motivation to discover a reachable path implemented by the scheme of UCA, while the efficiency is the second motivation to shorten the cruising distance. Meanwhile, the stabilized path solution is accomplished by hybrid-GA. In variable oceans, currents severely impact the path solution and lead the global optimum to absence. Therefore, alternative is to discover an optimal path with the minimum upstream-current sub-paths to approximate the minimal cruising distance in the condition that the discovered cruising distance should be less than the glider cruising range. To deeply analyze the path reachability, two theorems are developed to verify the conditions of the downstream-current angle (DCA). To evaluate the path-planning performances, 6 state-of-the-art fitness functions are studied and used to make a fair comparison with the EEF in hybrid-GA. First of all, 112 scenarios are created in the restricted random current variations (RRCV). Secondly, 21 scenarios are created in the near-real Kuroshio Current of east Taiwan (KCET) introducing from an ocean prediction model. These scenarios are designed to evaluate fairly the EEF in hybrid-GA. Numeric results show that whether the RRCV or the KCET, the proposed EEF indeed is able to discover the optimal path with the benefits of reachability and efficiency. Therefore, the proposed genetic-based effective approach is well developed to solve the GPP in variable oceans.

Online and offline based load balance algorithm in cloud computing

Abstract: Cloud Computing (CC) makes it possible for a common user to get an access to large pools of data and computational resources through a variety of interfaces. Among the so many important problems in CC, load balancing technique has been paid more and more attention for its important role. Good load balance algorithms can make whole system run more efficient. A new load balancing method combined with the advantage of online and offline load balancing algorithms are proposed in this paper. Two-choice algorithm and its improvement are used in the online step. Bacteria Foraging Optimization (BFO) and its improvement motivated by Lamarck Evolutionary Theory are introduced in our offline step. Online load balancing uses imperfect information, aiming at finishing tasks as fast as possible; while the offline makes full use of all information to make a supplement. Experiments on the heterogeneous tasks and serving points for computation intensive loads have been used here and the good performance shows the efficiency of our proposed method.

An Energy Balancing Strategy Based on Hilbert Curve and Genetic Algorithm for Wireless Sensor Networks

Abstract: A wireless sensor network is a sensing system composed of a few or thousands of sensor nodes. These nodes, however, are powered by internal batteries, which cannot be recharged or replaced, and have a limited lifespan. Traditional two-tier networks with one sink node are thus vulnerable to communication gaps caused by nodes dying when their battery power is depleted. In such cases, some nodes are disconnected with the sink node because intermediary nodes on the transmission path are dead. Energy load balancing is a technique for extending the lifespan of node batteries, thus preventing communication gaps and extending the network lifespan. However, while energy conservation is important, strategies that make the best use of available energy are also important. To decrease transmission energy cost and prolong network lifespan, a three-tier wireless sensor network is proposed, in which the first level is the sink node and the third-level nodes communicate with the sink node via the service sites on the second level. Moreover, this study aims to minimize the number of service sites to decrease the construction cost. Statistical evaluation criteria are used as benchmarks to compare traditional methods and the proposed method in the simulations.

A Matrix-Based Implementation of DE Algorithm: The Compensation and Deficiency

Abstract: Differential Evolution has become a very popular continuous optimization algorithm since its inception as its simplicity, easy coding and good performance over kinds of optimization problems Difference operator in donor vector calculation is the key feature of DE algorithm Usually, base vector and difference vectors selection in calculating a donor usually cost extra lines of condition judgement Moreover, these vectors are not equally selected from the individual population These lead to more perturbation in optimization performance To tackling this disadvantage of DE implementation, a matrix-based implementation of DE algorithm is advanced herein this paper Three commonly used DE implementation approaches in literature are also presented and contrasted CEC2013 test suites for real-parameter optimization are used as the test-beds for these comparison Experiment results show that the proposed matrix-based implementation of DE algorithm performs better on optimization performance than the common implementation schemes of DE algorithm with similar time complexity

QUATRE Algorithm with Sort Strategy for Global Optimization in Comparison with DE and PSO Variants

Abstract: Optimization algorithm in swarm intelligence is getting more and more prevalent both in theoretical field and in real-world applications. Many nature-inspired algorithms in this domain have been proposed and employed in different applications. In this paper, a new QUATRE algorithm with sort strategy is proposed for global optimization. QUATRE algorithm is a simple but powerful stochastic optimization algorithm proposed in 2016 and it tackles the representational/positional bias existing in DE structure. Here a sort strategy is used for the enhancement of the canonical QUATRE algorithm. This advancement is verified on CEC2013 test suite for real-parameter optimization and also is contrasted with several state-of-the-art algorithms including Particle Swarm Optimization (PSO) variants, Differential Evolution (DE) variants on COCO framework under BBOB2009 benchmarks. Experiment results show that the proposed QUATRE algorithm with sort strategy is competitive with the contrasted algorithms.

A Novel Hybrid GWO-FPA Algorithm for Optimization Applications

Abstract: The recent trend of research is to hybridize two or several numbers of variants to find out the better quality of solution in practical optimization applications. In this paper, a new approach hybrid Grey Wolf Optimizer (GWO)-Flower Pollination Algorithm (FPA) is proposed based on the combination of exploitation phase in GWO and exploration stage in FPA. The hybrid proposed GWOFPA improves movement directions and speed of the grey wolves in updating positions of FPA. The simulation uses six benchmark tests for evaluating the performance of the proposed method. Compared other metaheuristics such as Particle Swarm Optimization (PSO), FPA, and GWO, the simulation results demonstrate that the proposed approach offers the better performance in solving optimization problems with or without unknown search areas.

Deep convolutional neural networks-based age and gender classification with facial images

Abstract: In this paper, we build an age and gender classification system including two networks to classify age and gender based on GoogLeNet with the help of Caffe deep learning framework. It outputs gender and age groups of the facial images captured from the camera. We use Adience dataset to train GoogLeNet. Asynchronous Stochastic Gradient Descent based on multi-GPUs is used to optimize training process. We intend to use the trained network to build a classification system in real world to show the practicability. For instance, it can apply to a targeted delivery in bus stop or department store. The results indicate that the accuracy of the classification network can be improved by pre-training. In addition, the multi-GPUs training platform can improve the training speed during the recognition. Overall system reaches speed of 8fps with a high accuracy to classify age and gender.

An Unmanned Aerial Vehicle Optimal Route Planning Based on Compact Artificial Bee Colony

Abstract: In the inevitable trends of the modern aerial equipment, unmanned aerial vehicle (UAV) is one of the most concerned components to research. This paper presents a saving memory optimization algorithm of Compact artificial bee colony (cABC) for UAVs route planning problem. In the proposed method, route length and danger exposure are modeled mathematically as the objective function, and the compact algorithm concept is implemented to accommodate the route planning situation. In the compact algorithm, actual design variable of solutions search space of artificial bee colony algorithm is replaced with a probabilistic representation of the population. A probabilistic representation random of the collection behavior of bees is inspired to employ for this proposed algorithm. The real population is replaced with the probability vector updated based on single competition. The computational results compared with other algorithms in the literature shows that the proposed method can provide the effective way of using a modest memory for UAV route planning problem.

Deep learning application based on embedded GPU

Abstract: Since the Deep Neural Networks(DNNs) in the machine vision challenge has made outstanding achievements, a variety of excellent models are used in a variety of scenes. Embedded applications based on deep learning have become the key to becoming a product. This paper introduces the task of image recognition based on deep learning. And starts with the actual demand, it introduces the application of depth learning in commercial video analysis and accelerates the existing algorithm model. The results show that the optimized algorithm can realize the deep learning for embedded applications.

The QUasi-Affine TRansformation Evolution (QUATRE) Algorithm: An Overview

Abstract: QUasi-Affine TRansformation Evolution (QUATRE) algorithm is a new simple but powerful stochastic optimization algorithm proposed recently. The QUATRE algorithm aims to tackle the representational/positional bias inborn with DE algorithm and secures an overall better performance on commonly used Conference of Evolutionary Computation (CEC) benchmark functions. Recently, several QUATRE variants have been already proposed since its inception in 2016 and performed very well on many benchmark functions. In this paper, we mainly have a brief overview of all these proposed QUATRE variants first and then make simple contrasts between these QUATRE variants and several state-of-the-art DE variants under CEC2013 test suites for real-parameter single objective optimization benchmark functions. Experiment results show that the movement trajectory of individuals in the QUATRE structure is much more efficient than DE structure on most of the tested benchmark functions.

Reversible data hiding based on the local smoothness estimator and optional embedding strategy in four prediction modes

Abstract: Four new prediction modes are proposed in this paper, each of which is a three-step process for all to-be-embedded pixels (nearly three-fourths of all the pixels). By designing each mode reasonably, all to-be-embedded pixels can be predicted with high accuracy, and thus, the number of embeddable pixels can be increased largely. In each step, a local smoothness estimator is utilized to determine if one embeddable pixel is located in a smooth or complex region, which is defined as the variance of the total neighbors of this pixel. In fact, the correlation evaluated by using the total neighbors, instead of a part, can reflect the complexity of the region more accurately. In this paper, an optional embedding strategy is introduced so as to select a low-distortion reversible data hiding (RDH) method according to the desired embedding rate (ER). Specifically, when the required ER is low, difference expansion (DE) is used to process those pixels in smooth regions while leaving the rest unaltered. With ER largely increased, adaptive embedding is used to embed 2-bit into these pixels with low local variance by DE while 1-bit into the remaining ones. The experimental results also demonstrate the proposed method is effective.

On the Security of a Certificateless Public Key Encryption with Keyword Search

Abstract: Public key encryption with keyword search (PEKS) is one of searchable encryption mechanisms. It not only provides user to retrieve ciphertext by keyword but also protects the confidentiality of keyword. In the past, many PEKS schemes based on different cryptosystems were proposed. Recently. Zheng et al. proposed a certificateless based PEKS scheme called CLKS. In this paper, we show that Zheng et al.’s CLKS scheme has some security flaw, i.e. their scheme suffered from an off-line keyword guessing attack.

Scene text recognition algorithm based on faster RCNN

Abstract: Industrial session of the natural scene in the text recognition technology has a great demand. The traditional optical character recognition technology (OCR) requires the text neat layout and neatness and background clean, and industrial production often fail to meet such standards. In this paper, a new text recognition algorithm based on deep learning is proposed for the existing problems of OCR technology. In this paper, a new method based on convolution neural network (Faster RCNN) is proposed to improve the correctness of text recognition. Compared with the conventional detection method, the correct rate of recognition based on Faster RCNN model can reach 90.4%, and the correctness rate is 88.9%. Experiments show that the recognition method in this paper is effective.

Development of a Three-Dimensional Multimode Visual Immersive System With Applications in Telepresence

Abstract: Three dimensional (3-D) visual immersive system provides a better experience of virtual or remote environments compared with 2-D vision and is thus becoming one of the most active research fields of virtual reality. This paper describes how a 3-D multimode visual immersive system (3DMVIS) is developed, including system architecture, triplet lens module, head orientation tracking circuit, display module, and sensor signal processing. It supports two operation modes for different applications, corresponding to different combinations of lens. As a portable head-mounted display (HMD) device, it can be used in extensive virtual reality and telepresence applications. In addition, this HMD provides visual accommodation with the aid of manual focus and compatibility with various types of 3-D videos. Hence, users with various levels of myopia can experience this facility without wearing glasses. Meanwhile, a typical application in telepresence is investigated along with a compact panoramic camera system. The 3DMVIS is tested and evaluated with various virtual reality scenes, standard-format 3-D movies, and real-world 360° panoramic videos. In addition, comparisons with other commercial counterparts are presented. Technical specification indicates that the 3DMVIS with its telepresence system satisfies the requirements of practical applications.

Low-latency digit-serial dual basis multiplier for lightweight cryptosystems

Abstract: Various cryptosystems, such as elliptic curve and pairing-based cryptosystems, in resource-constrained security applications rely on finite field multiplication. For applications such as these, a digit-serial multiplier has the potential features to achieve a trade-off between space and time complexities. The authors propose an efficient decomposition of the multiplication into four independent sub-multiplication units to facilitate parallel processing, which is additionally facilitated by the systolic structures of the sub-multiplication units. The proposed architecture uses a four-bit scheme to construct a novel processing element, instead of using only one bit as is currently used in similar multipliers. The results of the synthesis show that the proposed digit-serial dual basis multiplier eliminates up to 96% of the critical path delay.

Remote sensing image object recognition based on convolutional neural network

Abstract: The development of space remote sensing technology brings a lot of remote sensing image data. The traditional target detection method is difficult to adapt to the large amount of high-resolution remote sensing image data. It is necessary to find a way to automatically learn the most effective features from the image data, and to fully recover the correlation between the data. Based on the recognition of the typical targets in remote sensing image data, this paper proposes a method of remote sensing target recognition based on deep learning. In this paper, the pre-train method is introduced to improve the simulation of the model. The experiment of the test set proves the validity of the method.

Optimal path planning for motion robots based on bees pollen optimization algorithm

Abstract: Due to interference phenomena among unnatural dimensions of the motion robots’ operations space, optimal path planning of them has to satisfy not just one criterion, but rather multi-objects. In this paper, we propose a novel multi-object approach for optimal mobile robot path planning, based on bees pollen optimizer (BPO). We consider two objects of distance and smooth path of the special plan for motion robots for constructing a minimization one. In operation environment for action robots, the location of the target and the obstacles are set up for the solution of BPO. The selected sequence of the mobile robot is a set of the chosen global best settlement in each iteration, which updates its archived data throughout the movement for motion robots in order. A series of simulations are executed in some environments for the best pathway once the robot reaches its goal. The results indicate that the proposed approach offered the robot path to its target without touching the obstacles, and the propos...

An Optimization Approach for Potential Power Generator Outputs Based on Parallelized Firefly Algorithm

Abstract: The parallel processing plays an important role in efficient and effective computations of function optimization. This paper proposes a parallel optimization algorithm for the committed generating electric powers of thermal plants based on Firefly Algorithm (FA). An economic condition for the power systems can be determined through the optimization techniques for transmission loss, power balance and generation capacity. The aim of the proposed parallel algorithm with communication strategies is to correlate agents in subgroups and to share the computation load over several machines. The expense criteria for each generation unit and the coefficient matrix are formulated as the objective function, which is to be computed in our parallel optimization for electricity flow of the transmission losses in the power systems. Four selected functions and two cases of six units and fifteen units of thermal plants are tested in our experiments for optimization. Through a comparison between different methods in related works, our experimental results show that the proposed method results in the higher effect and accuracy.

Novel Artificial Bee Colony Algorithm Based Load Balance Method In Cloud Computing.

Abstract: with rapid development of cloud computing and the Internet, load balance techniques are more and more important than ever. A good scheduling algorithm is an important method to solve the load balance problems. A new load balance algorithm based on ABC algorithm is proposed here. It can be seen as a new scheduling method based on swarm intelligence algorithm. Good experimental results proved by simulation tools called cloudsim has shown the efficiency.

Knowledge Graph Embedding Translation Based on Constraints.

Abstract: Knowledge graph reasoning is discovering new entity relations by computing and inference from existing relations. However, most reasoning models of translation embedding-based knowledge graphs have not considered the semantic-type constraints of relations in the construction of corrupted triplets. Hence, the constructed corrupted triplets may not conform to the actual semantic information and may, thus, significantly affect the prediction accuracy of the model. Therefore, we propose a constraint-based embedding model in this paper. First, the model establishes the head and tail entity set for each relationship. Then, it ensures that both the replacing head and tail entities in the corrupted triplet belong to the respective entity set so that the corrupted triplets that do not conform to the responding semantic relations are excluded. To evaluate the proposed model, we conduct link prediction and triple classification on WordNet and Freebase databases. The experimental results show that our method remarkably improves the performance compared to several state-of-the-art baselines.