To minimize the annotation costs associated with training semantic segmentation models and object detection models, weakly supervised detection and weakly supervised segmentation approaches have been extensively studied. However most of these approaches assume that the domain between training and testing is the same, which at times results in considerable performance drops. For example, if we train an object detection network using only web images showing a large object at the center, it can be difficult for the network to detect multiple small objects. In this paper, we focus on training a CNN with only web images and achieve object detection in the wild. A proposal-based approach can address the problem associated with differences in domains because web images are similar to images of the proposal. In both domains, the target object is located at the center of the image and the ratio of the size of the target object to the size of the image is large. Several proposal methods have been proposed to detect regions with high “object-ness.” However, many of these proposals generate a large number of candidates to increase the recall rate. Considering the recent advent of deep CNNs, methods that generate a large number of proposals exhibit problems in terms of processing time for practical use. Therefore, we propose a CNN-based “food-ness” proposal method in this paper that requires neither pixel-wise annotation nor bounding box annotation. Our method generates proposals through backpropagation and most of these proposals focus only on food objects. In addition, we can easily control the number of proposals. Through experiments, we trained a network model using only web images and tested the model on the UEC FOOD 100 dataset. We demonstrate that the proposed method achieves high performance compared to traditional proposal methods in terms of the trade-off between accuracy and computational cost. Therefore, in this paper, we propose an intermediate approach between the traditional proposal approach and the fully convolutional approach. In particular, we propose a novel proposal method that generates high“food-ness” regions using fully convolutional networks based on the backward approach by training food images gathered from the web.

In recent years, a rise in healthy eating has led to various food management applications which have image recognition function to record everyday meals automatically. However, most of the image recognition functions in the existing applications are not directly useful for multiple-dish food photos and cannot automatically estimate food calories. Meanwhile, methodologies on image recognition have advanced greatly because of the advent of Convolutional Neural Network (CNN). CNN has improved accuracies of various kinds of image recognition tasks such as classification and object detection. Therefore, we propose CNN-based food calorie estimation for multiple-dish food photos. Our method estimates dish locations and food calories simultaneously by multi-task learning of food dish detection and food calorie estimation with a single CNN. It is expected to achieve high speed and small network size by simultaneous estimation in a single network. Because currently there is no dataset of multiple-dish food photos annotated with both bounding boxes and food calories, in this work we use two types of datasets alternately for training a single CNN. For the two types of datasets, we use multiple-dish food photos annotated with bounding boxes and single-dish food photos with food calories. Our results showed that our multi-task method achieved higher accuracy, higher speed and smaller network size than a sequential model of food detection and food calorie estimation.

A novel configuration data compression technique for coarse-grained reconfigurable architectures (CGRAs) is proposed. Reducing the size of configuration data of CGRAs shortens the reconfiguration time especially when the communication bandwidth between a CGRA and a host CPU is limited. In addition, it saves energy consumption of configuration cache and controller. The proposed technique is based on a multicast configuration technique called RoMultiC, which reduces the configuration time by multicasting the same data to multiple PEs (Processing Elements) with two bit-maps. Scheduling algorithms for an optimizing the order of multicasting have been proposed. However, the multicasting is possible only if each PE has completely the same configuration. In general, configuration data for CGRAs can be divided into some fields like machine code formats of general perpose CPUs. The proposed scheme confines a part of fields for multicasting so that the possibility of multicasting more PEs can be increased. This paper analyzes algorithms to find a configuration pattern which maximizes the number of multicasted PEs. We implemented the proposed scheme to CMA (Cool Mega Array), a straight forward CGRA as a case study. Experimental results show that the proposed method achieves 40.0% smaller configuration than a previous method for an image processing application at maximum. The exploration of the multicasted grain size reveals the effective grain size for each algorithm. Furthermore, since both a dynamic power consumption of the configuration controller and a configuration time are improved, it achieves 50.1% reduction of the energy consumption for the configuration with a negligible area overhead.

This new design uses a low power embedded controller (EC) in cooperation with the BIOS of a notebook (NB) computer, both to accomplish dynamic adjustment and to maintain a required performance level of the battery mode of the notebook. In order to extend the operation time at the battery mode, in general, the notebook computer will directly reduce the clock rate and then reduce the performance. This design can obtain the necessary balance of the performance and the power consumption by using both the EC and the BIOS cooperatively to implement the dynamic control of both the CPU and the GPU frequency to maintain the system performance at a sufficient level for a high speed and high resolution video game. In contrast, in order to maintain a certain notebook performance, in terms of battery life it will be necessary to make some trade-offs.

Hardware prefetching involves a sophisticated balance between accuracy, coverage, and timeliness while minimizing hardware cost. Recent prefetchers have achieved these goals, but they still require complex hardware and a significant amount of storage. In this paper, we propose an efficient Per-page Most-Offset Prefetcher (PMOP) that minimizes hardware cost and simultaneously improves accuracy while maintaining coverage and timeliness. We achieve these objectives using an enhanced offset prefetcher that performs well with a reasonable hardware cost. Our approach first addresses coverage and timeliness by allowing multiple Most-Offset predictions. To minimize offset interference between pages, the PMOP leverages a fine-grain per-page offset filter. This filter records the access history with page-IDs, which enables efficient mapping and tracking of multiple offset streams from diverse pages. Analysis results show that PMOP outperforms the state-of-the-art Signature Path Prefetcher while reducing storage overhead by a factor of 3.4.

A concurrent system consists of multiple processes that are run simultaneously. The execution orders of these processes are defined by a scheduler. In model checking techniques, the scheduling policy is closely related to the search algorithm that explores all of the system states. To ensure the correctness of the system, the scheduling policy needs to be taken into account during the verification. Current approaches, which use fixed strategies, are only capable of limited kinds of policies and are difficult to extend to handle the variations of the schedulers. To address these problems, we propose a method using a domain-specific language (DSL) for the succinct specification of different scheduling policies. Necessary artifacts are automatically generated from the specification to analyze the behaviors of the system. We also propose a search algorithm for exploring the state space. Based on this method, we develop a tool to verify the system with the scheduler. Our experiments show that we could serve the variations of the schedulers easily and verify the systems accurately.

Compressed sensing is an effective compression algorithm. It is widely used to measure signals in distributed sensor networks (DSNs). Considering the limited resources of DSNs, the measurement matrices used in DSNs must be simple. In this paper, we construct a deterministic measurement matrix based on Gordon-Mills-Welch (GMW) sequence. The column vectors of the proposed measurement matrix are generated by cyclically shifting a GMW sequence. Compared with some state-of-the-art measurement matrices, the proposed measurement matrix has relative lower computational complexity and needs less storage space. It is suitable for resource-constrained DSNs. Moreover, because the proposed measurement matrix can be realized by using simple shift register, it is more practical. The simulation result shows that, in terms of recovery quality, the proposed measurement matrix performs better than some state-of-the-art measurement matrices.

The real-time state-of-health (SOH) estimation of lithium-ion batteries for electric vehicles (EV) is essential to EV maintenance. According to situations in practical applications such as long EV battery capacity test time, unavailability of regular daily tests, and availability of full-life-cycle charge data of EV recorded on the charging facility big data platform, this paper studies an online in-use EV state-of-health estimation method using iterated extended Gaussian process regression-Kalman filter (GPR-EKF) to incorporate lithium-ion battery data at the macro time scale and the micro time scale based on daily charge data of electric vehicles. This method proposes a kernel function GPR (Gaussian process regression) integrating neutral network with cycles to conduct fitting for data at the macro time scale to determine colored measurement noise; in addition, fragment charge data at the micro time scale is adjusted with real-time iteration to be used as the state equation, which effectively addresses issues of real-time SOC calibration and nonlinearization. The pertinence, effectiveness and real-time performance of the model algorithm in online battery state-of-health estimation is verified by actual data.

The 3D holographic display has long been expected as a future human interface as it does not require users to wear special devices. However, its heavy computation requirement prevents the realization of such displays. A recent study says that objects and holograms with several giga-pixels should be processed in real time for the realization of high resolution and wide view angle. To this problem, first, we have adapted a conventional FFT algorithm to a GPU cluster environment in order to avoid heavy inter-node communications. Then, we have applied several single-node and multi-node optimization and parallelization techniques. The single-node optimizations include a change of the way of object decomposition, reduction of data transfer between the CPU and GPU, kernel integration, stream processing, and utilization of multiple GPUs within a node. The multi-node optimizations include distribution methods of object data from host node to the other nodes. Experimental results show that intra-node optimizations attain 11.52 times speed-up from the original single node code. Further, multi-node optimizations using 8 nodes, 2 GPUs per node, attain an execution time of 4.28 sec for generating a 1.6 giga-pixel hologram from a 3.2 giga-pixel object. It means a 237.92 times speed-up of the sequential processing by CPU and 41.78 times speed-up of multi-threaded execution on multicore-CPU, using a conventional FFT-based algorithm.

Currently, visual perceptions generated by visual prosthesis are low resolution with unruly color and restricted grayscale. This severely restricts the ability of prosthetic implant to complete visual tasks in daily scenes. Some studies explore existing image processing techniques to improve the percepts of objects in prosthetic vision. However, most of them extract the moving objects and optimize the visual percepts in general dynamic scenes. The application of visual prosthesis in daily life scenes with high dynamic is greatly limited. Hence, in this study, a novel unsupervised moving object segmentation model is proposed to automatically extract the moving objects in high dynamic scene. In this model, foreground cues with spatiotemporal edge features and background cues with boundary-prior are exploited, the moving object proximity map are generated in dynamic scene according to the manifold ranking function. Moreover, the foreground and background cues are ranked simultaneously, and the moving objects are extracted by the two ranking maps integration. The evaluation experiment indicates that the proposed method can uniformly highlight the moving object and keep good boundaries in high dynamic scene with other methods. Based on this model, two optimization strategies are proposed to improve the perception of moving objects under simulated prosthetic vision. Experimental results demonstrate that the introduction of optimization strategies based on the moving object segmentation model can efficiently segment and enhance moving objects in high dynamic scene, and significantly improve the recognition performance of moving objects for the blind.

With advances in information technology and the development of big data, manual operation is unlikely to be a smart choice for stock market investing. Instead, the computer-based investment model is expected to bring investors more accurate strategic analysis and more effective investment decisions than human beings. This paper aims to improve investor profits by mining for critical information in the stock data, therefore helping big data analysis. We used the R language to find the technical indicators in the stock market, and then applied the technical indicators to the prediction. The proposed R package includes several analysis toolkits, such as trend line indicators, W type reversal patterns, V type reversal patterns, and the bull or bear market. The simulation results suggest that the developed R package can accurately present the tendency of the price and enhance the return on investment.

In this paper, a nonoverlapping multi-camera and people re-identification algorithm is proposed. It applies inflated major color features for re-identification to reduce computation time. The inflated major color features can dramatically improve efficiency while retaining high accuracy of object re-identification. The proposed method is evaluated over a wide range of experimental databases. The accuracy attains upwards of 40.7% in Rank 1 and 84% in Rank 10 on average, while it obtains three to 15 times faster than algorithms reported in the literature. The proposed algorithm has been implemented on a SOC-FPGA platform to reach 50 FPS with 1280×720 HD resolution and 25 FPS with 1920×1080 FHD resolution for real-time processing. The results show a performance improvement and reduction in computation complexity, which is especially ideal for embedded platform.

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.

This paper proposes a new rule-based approach to automatic question generation. The proposed approach focuses on analysis of both syntactic and semantic structure of a sentence. Although the primary objective of the designed system is question generation from sentences, automatic evaluation results shows that, it also achieves great performance on reading comprehension datasets, which focus on question generation from paragraphs. Especially, with respect to METEOR metric, the designed system significantly outperforms all other systems in automatic evaluation. As for human evaluation, the designed system exhibits similar performance by generating the most natural (human-like) questions.

In recent years, intelligent transportation system (ITS) techniques have been widely exploited to enhance the quality of public services. As one of the worldwide leaders in recycling, Taiwan adopts the waste collection and disposal policy named “trash doesn't touch the ground”, which requires the public to deliver garbage directly to the collection points for awaiting garbage collection. This study develops a travel time prediction system based on data clustering for providing real-time information on the arrival time of waste collection vehicle (WCV). The developed system consists of mobile devices (MDs), on-board units (OBUs), a fleet management server (FMS), and a data analysis server (DAS). A travel time prediction model utilizing the adaptive-based clustering technique coupled with a data feature selection procedure is devised and embedded in the DAS. While receiving inquiries from users' MDs and relevant data from WCVs' OBUs through the FMS, the DAS performs the devised model to yield the predicted arrival time of WCV. Our experiment result demonstrates that the proposed prediction model achieves an accuracy rate of 75.0% and outperforms the reference linear regression method and neural network technique, the accuracy rates of which are 14.7% and 27.6%, respectively. The developed system is effective as well as efficient and has gone online.

In this paper, we propose a novel framework for structural magnetic resonance image (sMRI) classification of Alzheimer's disease (AD) with data combination, outlier removal, and entropy-based data selection using AlexNet. In order to overcome problems of conventional classical machine learning methods, the AlexNet classifier, with a deep learning architecture, was employed for training and classification. A data permutation scheme including slice integration, outlier removal, and entropy-based sMRI slice selection is proposed to utilize the benefits of AlexNet. Experimental results show that the proposed framework can effectively utilize the AlexNet with the proposed data permutation scheme by significantly improving overall classification accuracies for AD classification. The proposed method achieves 95.35% and 98.74% classification accuracies on the OASIS and ADNI datasets, respectively, for the binary classification of AD and Normal Control (NC), and also achieves 98.06% accuracy for the ternary classification of AD, NC, and Mild Cognitive Impairment (MCI) on the ADNI dataset. The proposed method can attain significantly improved accuracy of up to 18.15%, compared to previously developed methods.

Machine learning algorithms are becoming more and more popular in current era. Data preprocessing especially feature selection is helpful for improving the performance of those algorithms. A new powerful feature selection algorithm is proposed. It combines the advantages of ant colony optimization and brain storm optimization which simulates the behavior of human beings. Six classical datasets and five state-of-art algorithms are used to make a comparison with our algorithm on binary classification problems. The results on accuracy, percent rate, recall rate, and F1 measures show that the developed algorithm is more excellent. Besides, it is no more complex than the compared approaches.

Redundant relations refer to explicit relations which can also be deducted implicitly. Although there exist several ontology redundancy elimination methods, they all do not take equivalent relations into consideration. Actually, real ontologies usually contain equivalent relations; their redundancies cannot be completely detected by existing algorithms. Aiming at solving this problem, this paper proposes a super-node based ontology redundancy elimination algorithm. The algorithm consists of super-node transformation and transitive redundancy elimination. During the super-node transformation process, nodes equivalent to each other are transferred into a super-node. Then by deleting the overlapped edges, redundancies relating to equivalent relations are eliminated. During the transitive redundancy elimination process, redundant relations are eliminated by comparing concept nodes' direct and indirect neighbors. Most notably, we proposed a theorem to validate real ontology's irredundancy. Our algorithm outperforms others on both real ontologies and synthetic dynamic ontologies.

Some of the most serious threats to network security involve malware. One common way to detect malware-infected machines in a network is by monitoring communications based on blacklists. However, such detection is problematic because (1) no blacklist is completely reliable, and (2) blacklists do not provide the sufficient evidence to allow administrators to determine the validity and accuracy of the detection results. In this paper, we propose a malicious DNS query clustering approach for blacklist-based detection. Unlike conventional classification, our cause-based classification can efficiently analyze malware communications, allowing infected machines in the network to be addressed swiftly.

In this letter, a novel frame aggregation scheduler is proposed to solve the head-of-line blocking problem for real-time user datagram protocol (UDP) traffic in error-prone and aggregation-enabled wireless local area networks (WLANs). The key to the proposed scheduler is to break the restriction of in-order delivery over the WLAN. The simulation results show that the proposed scheduler can achieve high UDP goodput and low delay compared to the conventional scheduler.

Temporal behavior is a primary aspect of business process executions. Herein, we propose a temporal outlier detection and analysis method for business processes. Particularly, the method performs correlation analysis between the execution times of traces and activities to determine the type of activities that significantly influences the anomalous temporal behavior of a trace. To this end, we describe the modeling of temporal behaviors considering different control-flow patterns of business processes. Further, an execution time matrix with execution times of activities in all traces is constructed by using the event logs. Based on this matrix, we perform temporal outlier detection and correlation-based analysis.

Over the past several years, the research of micro-expression recognition (MER) has become an active topic in affective computing and computer vision because of its potential value in many application fields, e.g., lie detection. However, most previous works assumed an ideal scenario that both training and testing samples belong to the same micro-expression database, which is easily broken in practice. In this letter, we hence consider a more challenging scenario that the training and testing samples come from different micro-expression databases and investigated unsupervised cross-database MER in which the source database is labeled while the label information of target database is entirely unseen. To solve this interesting problem, we propose an effective method called target-adapted least-squares regression (TALSR). The basic idea of TALSR is to learn a regression coefficient matrix based on the source samples and their provided label information and also enable this learned regression coefficient matrix to suit the target micro-expression database. We are thus able to use the learned regression coefficient matrix to predict the micro-expression categories of the target micro-expression samples. Extensive experiments on CASME II and SMIC micro-expression databases are conducted to evaluate the proposed TALSR. The experimental results show that our TALSR has better performance than lots of recent well-performing domain adaptation methods in dealing with unsupervised cross-database MER tasks.

In this research, we propose a novel method to determine fingerprint liveness to improve the discriminative behavior and classification accuracy of the combined features. This approach detects if a fingerprint is from a live or fake source. In this approach, fingerprint images are analyzed in the differential excitation (DE) component and the centralized binary pattern (CBP) component, which yield the DE image and CBP image, respectively. The images obtained are used to generate a two-dimensional histogram that is subsequently used as a feature vector. To decide if a fingerprint image is from a live or fake source, the feature vector is processed using support vector machine (SVM) classifiers. To evaluate the performance of the proposed method and compare it to existing approaches, we conducted experiments using the datasets from the 2011 and 2015 Liveness Detection Competition (LivDet), collected from four sensors. The results show that the proposed method gave comparable or even better results and further prove that methods derived from combination of features provide a better performance than existing methods.

Despite the widespread use of deep learning for speech emotion recognition, they are severely restricted due to the information loss in the high layer of deep neural networks, as well as the degradation problem. In order to efficiently utilize information and solve degradation, attention-based dense long short-term memory (LSTM) is proposed for speech emotion recognition. LSTM networks with the ability to process time series such as speech are constructed into which attention-based dense connections are introduced. That means the weight coefficients are added to skip-connections of each layer to distinguish the difference of the emotional information between layers and avoid the interference of redundant information from the bottom layer to the effective information from the top layer. The experiments demonstrate that proposed method improves the recognition performance by 12% and 7% on eNTERFACE and IEMOCAP corpus respectively.

This letter presents an objective quality index for benchmarking image inpainting algorithms. Under the guidance of the masks of damaged areas, the boundary region and the inpainting region are first located. Then, the statistical features are extracted from the boundary and inpainting regions respectively. For the boundary region, we utilize Weibull distribution to fit the gradient magnitude histograms of the exterior and interior regions around the boundary, and the Kullback-Leibler Divergence (KLD) is calculated to measure the boundary distortions caused by imperfect inpainting. Meanwhile, the quality of the inpainting region is measured by comparing the naturalness factors between the inpainted image and the reference image. Experimental results demonstrate that the proposed metric outperforms the relevant state-of-the-art quality metrics.

Background subtraction algorithms generate a background model of the monitoring scene and compare the background model with the current video frame to detect foreground objects. In general, most of the background subtraction algorithms fail to detect foreground objects when the scene illumination changes. An entropy based background subtraction algorithm is proposed to address this problem. The proposed method adapts to illumination changes by updating the background model according to differences in entropy value between the current frame and the previous frame. This entropy based background modeling can efficiently handle both sudden and gradual illumination variations. The proposed algorithm is tested in six video sequences and compared with four algorithms to demonstrate its efficiency in terms of F-score, similarity and frame rate.