This PDF file contains the front matter associated with SPIE Proceedings Volume 10011, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.

For the problem of ignoring color information and computing complexity and so on, a new target tracking algorithm based on improved SURF(Speed Up Robust Features) algorithm and Kalman filter fusion is studied. First, the color invariants are added in the generation process of SURF. And then the current position is predicted by using the Kalman filter and establishing the search window. Finally, the feature vectors in the search window are extracted by using the improved SURF algorithm for matching. The experiments prove that the algorithm can always track targets stably when the target appears scale changed, rotation and partial occlusion, and the tracking speed is greatly improved than that of the SURF algorithm.

In the field of intelligent transportation system a great number of vision-based techniques have been proposed to prevent pedestrians from being hit by vehicles. This paper presents a system that can perform pedestrian and vehicle detection and monitoring of illegal activity in zebra crossings. In zebra crossing, according to the traffic light status, to fully avoid a collision, a driver or pedestrian should be warned earlier if they possess any illegal moves. In this research, at first, we detect the traffic light status of pedestrian and monitor the crossroad for vehicle pedestrian moves. The background subtraction based object detection and tracking is performed to detect pedestrian and vehicles in crossroads. Shadow removal, blob segmentation, trajectory analysis etc. are used to improve the object detection and classification performance. We demonstrate the experiment in several video sequences which are recorded in different time and environment such as day time and night time, sunny and raining environment. Our experimental results show that such simple and efficient technique can be used successfully as a traffic surveillance system to prevent accidents in zebra crossings.

This paper is devoted to the development of the software for online log deck volume measurement via photometry. Log volume measurement is based on the analysis of its cut which specific feature is rounded shape. Thus the development of the rounded object automatic detection method has to be the main part of the R&D. It was decided to use the fast radial symmetry algorithm as a basis of the detection method. Then its modification according to the specific of the application scope was implemented. The obtained method was introduced into the developed software. Analysis of its testing on 700 images gives the average probability of target object detection at 95,7% with lower range at 86%.

This paper proposes a new technique for rectangle detection in images based on hierarchical feature complexity. The algorithm follows a bottom-up/top-down approach: in the bottom-up phase, contour curves are extracted and its edges are fit to straight lines. Long contours may grow away from the object boundary and they may not complete a loop due to missing edges. The proposed algorithm introduces a solution to such problems in the top-down phase through two simple rules. First, contours are split into segments at the point where non-convexity occurs since this is the point where long contours depart from the object boundary. Second, the split segments are classified into six classes according to their probability of being a rectangle depending on the numbers of the segment sides and right angles they enclose. These classes are then completed into rectangles by searching for suitable lines that may have been missed during the bottom-up phase. The method is verified through experiments on a set of images covering several applications. The results are compared to state of the art methods and benchmarked to groundtruth.

At present, the detection and tracking for video moving object have been used in many fields. Aimed at the limitation of the traditional adjacent frame difference method, this paper presented the three-frame difference method to detect objects. For moving object tracking, this paper proposed a method combining Kalman filter and Mean-Shift, and used the prediction function of Kalman filter to overcome the defect of Mean-Shift in selecting the initial position of the candidate object. Experimental results showed that the detection and tracking method proposed in this paper are simple, precise, and perform a good result.

This paper studies the impact of margin distribution on detection performance and proposes Diverse Margin Distribution Ensemble (DMDE) for pedestrian detection, based on HOG descriptor. Large margin Distribution Machine (LDM) introduces the margin mean and margin variance. Large margin mean is relevant to the strong generalization performance and large margin variance is relevant to the more balanced detection rate between two classes. Inspired by this recognition, DMDE is proposed to obtain greater robustness and balance for pedestrian detection. It is a blending of SVM and two LDMs with different parameter orders and can aggregate the merits of the three classifiers. Experimental results show that DMDE is more robust and balanced than single SVM or LDM for pedestrian detection.

Histogram of Oriented Gradient (HOG) proposed by Dalal and Triggs is currently the most basic algorithm to detection pedestrian. The algorithm is weak to occlusion, since the algorithm trained by the image of pedestrian full body images as one feature. As a result, the detection rate using HOG feature becomes decreases remarkably. To solve this problem, the paper proposed detection system using Deformable Part-based Model (DPM) just divided two parts of pedestrian data through latent Support Vector Machine (SVM) based machine learning. Experimental results show that proposed approach achieves better performance on detection with high accuracy than existed method [1].

In this paper, a new texture descriptor based on the extraction of image representation which is the selection of weights assigned to input and output of local binary patterns in determining the efficiency of each feature, called weight-incorporated local binary pattern (WiLBP), is developed for image representation. By using averaged gradients information, the principal components of a covariance matrix are derived to obtain an adjusted principal components of a maximum variance matrix, namely quantized eigen-analysis (QEA). The QEA matrix is a weight matrix used to adjust the contribution of comparisons of pixel intensities. To evaluate the performance of the WiLBP, a series of experiments was tested on some popular face databases. The misclassification error obtained by the QEA across most trials is lower than that of the PCA. The experimental results also show that the WiLBP is a fast and robust method in individual recognition and gender classification applications.

This paper presents an approach to pain expression classification based on Gabor energy filters with Support Vector Machines (SVMs), followed by analyzing the effects of training set variations on the systems classification rate. This approach is tested on the UNBC-McMaster Shoulder Pain Archive, which consists of spontaneous pain images, hand labelled using the Prkachin and Solomon Pain Intensity scale. In this paper, the subjects pain intensity level has been quantized into three disjoint groups: no pain, weak pain and strong pain. The results of experiments show that Gabor energy filters with SVMs provide comparable or better results to previous filter- based pain recognition methods, with precision rates of 74%, 30% and 78% for no pain, weak pain and strong pain, respectively. The study of effects of intra-class skew, or changing the number of images per subject, show that both completely removing and over-representing poor quality subjects in the training set has little effect on the overall accuracy of the system. This result suggests that poor quality subjects could be removed from the training set to save offline training time and that SVM is robust not only to outliers in training data, but also to significant amounts of poor quality data mixed into the training sets.

Identification photo is a category of facial image that has strict requirements on image quality like size, illumination, user expression, dressing, etc. Traditionally, these photos are taken in professional studios. With the rapid popularity of mobile devices, how to conveniently take identification photo at any time and anywhere with such devices is an interesting problem. In this paper, we propose a novel semi-automatic identification photo generation approach. Given a user image, facial pose and expression are first normalized to meet the basic requirements. To correct uneven lighting condition in photo, an facial illumination normalization approach is adopted to further improve the image quality. Finally, foreground user is extracted and re-targeted to a specific photo size. Besides, background can also be changed as required. Preliminary experimental results show that the proposed method is efficient and effective in identification photo generation compared to commercial software based manual tunning.

With the growth of biometric technology, spoofing attacks have been emerged a threat to the security of the system. Main spoofing scenarios in the face recognition system include the printing attack, replay attack, and 3D mask attack. To prevent such attacks, techniques that evaluating liveness of the biometric data can be considered as a solution. In this paper, a novel face liveness detection method based on cardiac signal extracted from face is presented. The key point of proposed method is that the cardiac characteristic is detected in live faces but not detected in non-live faces. Experimental results showed that the proposed method can be effective way for determining printing attack or 3D mask attack.

This paper proposes a face verification system that runs on Android mobile devices. In this system, facial image is captured by a built-in camera on the Android device firstly, and then face detection is implemented using Haar-like features and AdaBoost learning algorithm. The proposed system verify the detected face using histogram based features, which are generated by binary Vector Quantization (VQ) histogram using DCT coefficients in low frequency domains, as well as Improved Local Binary Pattern (Improved LBP) histogram in spatial domain. Verification results with different type of histogram based features are first obtained separately and then combined by weighted averaging. We evaluate our proposed algorithm by using publicly available ORL database and facial images captured by an Android tablet.

Visual experience of surface properties relies on accurately attributing encoded luminance variations (e.g., edges and contours) to any one of several potential environmental causes. We examined the role of differences in the local shading directions across sharp contours in (i) identifying occlusion boundaries and (ii) perceiving the depth layout of adjacent surfaces. We used graphical rendering to control the orientation of a simulated light source, and hence the shading direction between adjacent surface regions that met at a common edge. We call the difference in shading direction across the edge the delta shading angle. We found that delta-shaded edges looked like occluding boundaries. We also found that the perceived figure-ground organisation of the adjacent surface regions depended on an assumed lighting from above prior. Shaded regions experienced as convex surfaces illuminated from above were perceived as occluding surfaces in the foreground. We computed an image-based measure of delta shading using the difference in local shading direction (the orientation field) and found this model could reliably account for observer judgments of surface occlusion, better than local (in-)coherence in the orientation of isophotes across the edge alone. However, additional information from co-alignment of isophotes relative to the edge is necessary to explain figure-ground distinctions across a broad class of occlusion events. We conclude that both local and global measures of shading direction are needed to explain perceived scene organisation, and material appearance more generally.

This paper presents a concept of intercepting a moving table tennis ball using a robot. The robot has four degrees of freedom(DOF) which are simplified in such a way that The system is able to perform the task within the bounded limit. It employs computer vision to localize the ball. For ball identification, Colour Based Threshold Segmentation(CBTS) and Background Subtraction(BS) methodologies are used. Coordinate Transformation(CT) is employed to transform the data, which is taken based on camera coordinate frame to the general coordinate frame. The sensory system consisted of two HD Web Cameras. The computation time of image processing from web cameras is long .it is not possible to intercept table tennis ball using only image processing. Therefore the projectile motion model is employed to predict the final destination of the ball.

Motion segmentation is a fundamental step for vehicle detection especially in urban traffic surveillance systems. Temporal frame differencing is the simplest and fastest technique that is used to identify foreground moving vehicles from static background scene. Conventional techniques utilize background modelling and subtraction, which involves poor adaptation under slow or temporarily stopped vehicles. To address this problems cumulative frame differencing (CFD) is proposed. Dynamic threshold value based on the standard deviation of CFD is used to estimate global variance of the motion accumulated variations of pixel intensity. The tests of the proposed technique achieve robust and accurate vehicle segmentation, which improves detection of slow motion, temporary and long term stopped vehicles, moreover, it enables the real-time capability.

High efficiency video coding (HEVC) is a video compression standard that outperforms the predecessor H.264/AVC by doubling the compression efficiency. To enhance the compression accuracy, the partition sizes ranging is from 4x4 to 64x64 in HEVC. However, the manifold partition sizes dramatically increase the encoding complexity. This paper proposes a fast depth decision based on spatial and temporal correlation. Spatial correlation utilize the code tree unit (CTU) Splitting information and temporal correlation utilize the motion vector predictor represented CTU in inter prediction to determine the maximum depth in each CTU. Experimental results show that the proposed method saves about 29.1% of the original processing time with 0.9% of BD-bitrate increase on average.

The article represents the results of applying various machine learning methods for solving the problem of identifying the edges of a left ventricle area in ultrasound images. The problem of the determination this area consider as a special case of binary classification of pixels. It is shown that the level of correct classification of pixels more than 94.6%. The complex model Bag-10 demonstrates the best result of classification - 98.4%.

In ISO TC249 conference, tongue diagnosis has been one of the most active research and their objectifications has become significant with the help of numerous statistical and machine learning algorithm. Color information of substance or tongue body has kept valuable information regarding the state of disease and its correlation with the internal organs. In order to produce high reproducibility of color measurement analysis, tongue images have to undergo several procedures such as color correction, segmentation and tongue’s substance-coating separation. This paper presents a novel method to recognize substance and coating from tongue images and eliminate the tongue coating for accurate substance color measurement for diagnosis. By utilizing Hue, Saturation, Value (HSV) color space, new color-brightness threshold parameters have been devised to improve the efficiency of tongue’s substance and coating separation procedures and eliminate shadows. The algorithm offers fast processing time around 0.98 seconds for 60,000 pixels tongue image. The successful tongue’s substance and coating separation rate reported is 90% compared to the labelled data verified by the practitioners. Using 300 tongue images, the substance Lab color measurement with small standard deviation had revealed the effectiveness of this proposed method in computerized tongue diagnosis system.

Identification of Dendritic Cell (DC) particularly in the cancer microenvironment is a unique disclosure since fighting tumor from the harnessing immune system has been a novel treatment under investigation. Nowadays, the staining procedure in sorting DC can affect their viability. In this paper, a computer aided system is proposed for automatic classification of DC in peripheral blood mononuclear cell (PBMC) images. Initially, the images undergo a few steps in preprocessing to remove uneven illumination and artifacts around the cells. In segmentation, morphological operators and Canny edge are implemented to isolate the cell shapes and extract the contours. Following that, information from the contours are extracted based on Fourier descriptors, derived from one dimensional (1D) shape signatures. Eventually, cells are classified as DC by comparing template matching (TM) of established template and target images. The results show that the proposed scheme is reliable and effective to recognize DC.

The purpose of this paper is to provide an in-depth analysis of computer aided system for the early diagnosis of Deep Vein Thrombosis (DVT). Normally, patients are diagnosed with DVT through ultrasound examination after they have a serious complication. Thus, this study proposes a new approach to reduce the risk of recurrent DVT by tracking the venous valve movement behaviour. Inspired by image processing technology, several image processing methods namely, image enhancement, segmentation and morphological have been implemented to improve the image quality for further tracking procedure. In segmentation, Otsu thresholding provides a significant result in segmenting valve structure. Subsequently, morphological dilation method is able to enhance the region shape of the valve distinctly and precisely. Lastly, image subtraction method is presented and evaluated to track the valve movement. Based on the experimental results the normal range of valve velocity lies within the range of blood flow velocity (V_b) and occasionally may result in higher values.

Epilepsy is a neurological disorder which can, if not controlled, potentially cause unexpected death. It is extremely crucial to have accurate automatic pattern recognition and data mining techniques to detect the onset of seizures and inform care-givers to help the patients. EEG signals are the preferred biosignals for diagnosis of epileptic patients. Most of the existing pattern recognition techniques used in EEG analysis leverage the notion of supervised machine learning algorithms. Since seizure data are heavily under-represented, such techniques are not always practical particularly when the labeled data is not sufficiently available or when disease progression is rapid and the corresponding EEG footprint pattern will not be robust. Furthermore, EEG pattern change is highly individual dependent and requires experienced specialists to annotate the seizure and non-seizure events. In this work, we present an unsupervised technique to discriminate seizures and non-seizures events. We employ power spectral density of EEG signals in different frequency bands that are informative features to accurately cluster seizure and non-seizure events. The experimental results tried so far indicate achieving more than 90% accuracy in clustering seizure and non-seizure events without having any prior knowledge on patient's history.

Our paper presents a method for reconstructing a high-resolution (HR) image from a set of multi-view color images captured by a camera array. First, an accurate depth map of low-resolution (LR) image captured by a selected reference camera is obtained using graph cuts. Then, a HR image corresponding to the reference camera can be estimated by super-resolution reconstruction. Experiments on real images show the effectiveness of our method.

The state-of-the-art blind image deblurring (BID) methods are sensitive to noise, and most of them can deal with only small levels of Gaussian noise. In this paper, we use simple filters to present a robust BID framework which is able to robustify exiting BID methods to high-level Gaussian noise or/and Non-Gaussian noise. Experiments on images in presence of Gaussian noise, impulse noise (salt-and-pepper noise and random-valued noise) and mixed Gaussian-impulse noise, and a real-world blurry and noisy image show that the proposed method can faster estimate sharper kernels and better images, than that obtained by other methods.

High dynamic range (HDR) images can show more details and luminance information in general display device than low dynamic image (LDR) images. We present a robust HDR imaging system which can deal with blurry LDR images, overcoming the limitations of most existing HDR methods. Experiments on real images show the effectiveness and competitiveness of the proposed method.

We propose an image noise removal method based on spatial gradient based bilateral filter and smooth filtering. Our method consist two step process; in first step, for a given noisy image we extract all of its patches and apply our newly developed spatial gradient based bilateral filter on each patch and get an reference image; in second step we perform smooth filtering on each pixel of the reference image. Experimental results show that our method is consistent and comparable or better than state-of-the-art.

This paper proposes an efficient method to improve visibility of dark color images and video sequences. Visibility improvement of dark color images play a significant role in computer vision, digital image processing and pattern recognition. In the proposed method we have worked in hue-saturation-value (HSV) color model. The proposed method initially decomposes the V-plane of the input image into low and high frequency components using DCT; it then estimates the singular value matrix of the low-frequency component. After applying certain processing to improve visibility of the dark color image, it reconstructs the processed image by applying inverse DCT. Experimental results show that the proposed method produces enhanced images of comparable or higher quality than those produced using previous state-of-the-art methods.

The feature descriptors such as SIFT (Scale Invariant Feature Transform), SURF (Speeded-up Robust Features) and ORB (Oriented FAST and Rotated BRIEF) are known as the most commonly used solutions for the content-based image retrieval problems. In this paper, a novel approach called ”Weighted Feature Fusion” is proposed as a generic solution instead of applying problem-specific descriptors alone. Experiments were performed on two basic data sets of the Inria in order to improve the precision of retrieval results. It was found that in cases where the descriptors were used alone the proposed approach yielded 10-30% more accurate results than the ORB alone. Besides, it yielded 9-22% and 12-29% less False Positives compared to the SIFT alone and SURF alone, respectively.

In this paper, we propose an air-touch interaction system for the tabletop type integral imaging 3D display. This system consists of the real 3D image generation system based on integral imaging technique and the interaction device using a real-time finger detection interface. In this system, we used multi-layer B-spline surface approximation to detect the fingertip and gesture easily in less than 10cm height from the screen via input the hand image. The proposed system can be used in effective human computer interaction method for the tabletop type 3D display.

3D(Three-Dimensional) point cloud registration technology is the hot topic in the field of 3D reconstruction, but most of the registration method is not real-time and ineffective. This paper proposes a point cloud registration method of 3D reconstruction based on Harris-SIFT and assistant camera. The assistant camera is used to pinpoint mobile 3D reconstruction device, The feature points of images are detected by using Harris operator, the main orientation for each feature point is calculated, and lastly, the feature point descriptors are generated after rotating the coordinates of the descriptors relative to the feature points’ main orientations. Experimental results of demonstrate the effectiveness of the proposed method.

Currently mobile apps for document scanning do not provide convenient operations to tackle large-size documents. In this paper, we present a one-click scanning approach for large-size documents using mobile phone camera. After capturing a continuous video of documents, our approach automatically extracts several key frames by optical flow analysis. Then based on key frames, a mobile GPU based image stitching method is adopted to generate a completed document image with high details. There are no extra manual intervention in the process and experimental results show that our app performs well, showing convenience and practicability for daily life.

With the popular use of high resolution remote sensing (HRRS) satellite images, a huge research efforts have been placed on change detection (CD) problem. An effective feature selection method can significantly boost the final result. While hand-designed features have proven difficulties to design features that effectively capture high and mid-level representations, the recent developments in machine learning (Deep Learning) omit this problem by learning hierarchical representation in an unsupervised manner directly from data without human intervention. In this letter, we propose approaching the change detection problem from a feature learning perspective. A novel deep Convolutional Neural Networks (CNN) features based HR satellite images change detection method is proposed. The main guideline is to produce a change detection map directly from two images using a pretrained CNN. This method can omit the limited performance of hand-crafted features. Firstly, CNN features are extracted through different convolutional layers. Then, a concatenation step is evaluated after an normalization step, resulting in a unique higher dimensional feature map. Finally, a change map was computed using pixel-wise Euclidean distance. Our method has been validated on real bitemporal HRRS satellite images according to qualitative and quantitative analyses. The results obtained confirm the interest of the proposed method.

A roll of steel might have various defects of scratch, stains, and chisel mark after slitting process. However, the traditional steel surface inspection method is via the human inspection that not only takes amount of time but also causes inconsistent inspection consequences. As a result, this paper proposed an in-line visual inspection hardware and software system. The hardware is composed of upper and lower optical module. The defect inspection algorithm includes automatic region of interesting (ROI) searching and defect detection by using Sobel method. Experimentations revealed that the successful detection rate is up to 80% and the inspection speed of per image with 3K in width and 1K in length is less than 80 milliseconds. The contribution is that the proposed method can provide suitable inspection results of the steel surface defect and meet the steel industry demands.

The future information systems are expected to be more intelligent and will take human queries in natural language as input and answer them promptly. To develop a chatbot or a computer program that can chat with humans in realistic manner to extent that human get impressions that he/she is talking with other human is a challenging task. To make such chatbots, different technologies will work together ranging from artificial intelligence to development of semantic resources. Sophisticated chatbots are developed to perform conversation in number of languages. Arabic chatbots can be helpful in automating many operations and serve people who only know Arabic language. However, the technology for Arabic language is still in its infancy stage due to some challenges surrounding the Arabic language. This paper offers an overview of the chatbot application and the several obstacles and challenges that need to be resolved to develop an effective Arabic chatbot.

For intelligent service robots, indoor scene classification is an important issue. To overcome the weak real-time performance of conventional algorithms, a new method based on Cloud computing is proposed for global image features in indoor scene classification. With MapReduce method, global PHOG feature of indoor scene image is extracted in parallel. And, feature eigenvector is used to train the decision classifier through SVM concurrently. Then, the indoor scene is validly classified by decision classifier. To verify the algorithm performance, we carried out an experiment with 350 typical indoor scene images from MIT LabelMe image library. Experimental results show that the proposed algorithm can attain better real-time performance. Generally, it is 1.4∼2.1 times faster than traditional classification methods which rely on single computation, while keeping stable classification correct rate as 70%.

There are many development of intelligent service robot in order to interact with user naturally. This purpose can be done by embedding speech and face recognition ability on specific tasks to the robot. In this research, we would like to propose Intelligent Coffee Maker Robot which the speech recognition is based on Indonesian language and powered by statistical dialogue systems. This kind of robot can be used in the office, supermarket or restaurant. In our scenario, robot will recognize user’s face and then accept commands from the user to do an action, specifically in making a coffee. Based on our previous work, the accuracy for speech recognition is about 86% and face recognition is about 93% in laboratory experiments. The main problem in here is to know the intention of user about how sweetness of the coffee. The intelligent coffee maker robot should conclude the user intention through conversation under unreliable automatic speech in noisy environment. In this paper, this spoken dialog problem is treated as a partially observable Markov decision process (POMDP). We describe how this formulation establish a promising framework by empirical results. The dialog simulations are presented which demonstrate significant quantitative outcome.

Safety is a vital issue in Human-Robot Interaction (HRI). In order to guarantee safety in HRI, a model reference impedance control can be a very useful approach introducing a compliant control. In particular, this paper establishes a fuzzy logic compliance control (i.e. active compliance control) to reduce impact and forces during physical interaction between humans/objects and robots. Exploiting a virtual mass-spring-damper system allows us to determine a desired compliant level by understanding the behavior of the model reference impedance control. The performance of fuzzy logic compliant control is tested in simulation for a robotic hand known as the RED Hand. The results show that the fuzzy logic is a feasible control approach, particularly to control position and to provide compliant control. In addition, the fuzzy logic control allows us to simplify the controller design process (i.e. avoid complex computation) when dealing with nonlinearities and uncertainties.

Force Sensing Resistors (FSR) sensors are devices that allow measuring static and dynamic forces applied to a contact surface. Their range of responses is basically depending on the variation of its electric resistance. In general, Flexiforce and Interlink are two common types of FSR sensors that are available, cheap and easily found in the market. Studies have shown that the FSR sensors are usually applied for robotic grippers and for biomechanical fields. This paper provides a brief overview of the application of the FSR sensors. Subsequently, two different set of experiments are carried out to test the effectiveness of the Flexiforce and Interlink sensors. First, the hardness detector system (Case Study A) and second, the force-position control system (Case Study B). The hardware used for the experiment was developed from low-cost materials. The results revealed that both FSR sensors are sufficient and reliable to provide a good sensing modality particularly for measuring force. Apart from the low-cost sensors, essentially, the FSR sensors are very useful devices that able to provide a good active compliance control, particularly for the grasping robotic hand.

Nowadays, there are many developments in building intelligent humanoid robot, mainly in order to handle voice and image. In this research, we propose blind speech separation system using FastICA for audio filtering and separation that can be used in education or entertainment. Our main problem is to separate the multi speech sources and also to filter irrelevant noises. After speech separation step, the results will be integrated with our previous speech and face recognition system which is based on Bioloid GP robot and Raspberry Pi 2 as controller. The experimental results show the accuracy of our blind speech separation system is about 88% in command and query recognition cases.

This paper is to improve the stability issue of the bipedal walking robot. The study of robot’s pivot joint constructs the driver system to control the implementation. First, a Proportion-Integral-Derivative (PID) controller is designed by which is used the concept of tuning parameter to achieve the stability of the system. Second, Fuzzy controller and tradition PID controller is used to maintain output. It improved original PID controller efficacy. Finally, Artificial Neuro-Fuzzy Inference System (ANFIS) is utilized which is made the controller to achieve self-studying and modify the effect which is completed by the intelligent controller. It improved bipedal robot’s stability control of realization. The result is verified that the walking stability of the bipedal walking robot in Matlab/Simulink. The intelligent controller has achieved the desired position of motor joint and the target stability performance.

Picking up objects in arbitrary poses is an important step in manufacturing. However, the occlusion of object will cause the picking process difficult. This paper presents a hierarchical detection method to estimate the pose of the object such as rod and bearing even in occluding. Combining the ellipse detection with the template matching, it is possible to identify the pose of a target object that is not be occluded. The propose method will enable a robot to grasp a non-occluded object and ensure a successful picking. Experiments witness the validity the method.

Existing packaging systems rely on human operation to position a box in the packaging device perform do the packaging task. Current facilities are not capable of handling boxes with different sizes in a flexible way. In order to improve the above-mentioned problems, an eye-to-hand visual servo automated packaging approach is proposed in this paper. The system employs two cameras to observe the box and the gripper mounted on the robotic manipulator to precisely control the manipulator to complete the packaging task. The system first employs two-camera vision to determine the box pose. With appropriate task encoding, a closed-loop visual servoing controller is designed to drive a manipulator to accomplish packaging tasks. The proposed approach can be used to complete automated packaging tasks in the case of uncertain location and size of the box. The system has been successfully validated by experimenting with an industrial robotic manipulator for postal box packaging.

Chord transcription is valuable to do by itself. It is known that the manual transcription of chords is very tiresome, time-consuming. It requires, moreover, musical knowledge. Automatic chord recognition has recently attracted a number of researches in the Music Information Retrieval field. It has known that a pitch class profile (PCP) is the commonly signal representation of musical harmonic analysis. However, the PCP may contain additional non-harmonic noise such as harmonic overtones and transient noise. The problem of non-harmonic might be generating the sound energy in term of frequency more than the actual notes of the respective chord. Autoencoder neural network may be trained to learn a mapping from low level feature to one or more higher-level representation. These high-level representations can explain dependencies of the inputs and reduce the effect of non-harmonic noise. Then these improve features are fed into neural network classifier. The proposed high-level musical features show 80.90% of accuracy. The experimental results have shown that the proposed approach can achieve better performance in comparison with other based method.

Aiming at the problem of separating the useful signal in the chaos background and using the phase matching method, the signal can be extracted effectively from the chaotic background. In this method, the chaotic background is not estimated with phase reconstruction and the geometric analysis of phase space is not required. Through the separation Simulation of the sinusoidal signal in the chaos background and the separation degree analysis, the low signal to noise ratio of the signal in the chaos background can be effectively separated. The effect of removing the chaotic background noise is obvious.

This paper presents a combination approach which fusing the estimates of forward backward pursuit (FBP) and backtracking-based adaptive orthogonal matching pursuit (BAOMP) to approximate sparse solutions for compressed sensing without the sparsity level as a prior. This algorithm referred to as combination approach for compressed sensing (CACS). It can improve the sparse signal recovery performance in a minimum number of measurements. Numerical experiments for both synthetic and real signals are conducted to demonstrate the validity and high performance of the proposed algorithm, as compared to the individual compressed sensing algorithms.

This study presents a new method to combine ECG biometrics with data compression within a common JPEG2000 framework. We target the two-lead ECG configuration that is routinely used in long-term heart monitoring. Incorporation of compressed-domain biometric techniques enables faster person identification as it by-passes the full decompression. Experiments on public ECG databases demonstrate the validity of the proposed method for biometric identification with high accuracies on both healthy and diseased subjects.

Publisher's Note: This paper, originally published on 7/11/16, was replaced with a corrected/revised version on 10/17/18. If you downloaded the original PDF but are unable to access the revision, please contact SPIE Digital Library Customer Service for assistance.

In this paper a low-cost solution for surface EMG (sEMG) signal retrieval is presented. The principal goal is to enable reading the temporal parameters of muscles activity by a computer device, with its further processing. Paper integrates design and deployment of surface electrodes and amplifier following the prior researches. Bearing in mind the goal of creating low-cost solution, the Arduino micro-controller was utilized for analog-to-digital conversion and communication. The software part of the system employs support vector machine (SVM) to classify the EMG signal, as acquired from sensors. Accuracy of the proposed solution achieves over 90 percent for six hand movements. Proposed solution is to be tested as an assistive device for several cases, involving people with motor disabilities and amputees.

Sound-event classification often extracts features from an image-like spectrogram. Recent approaches such as spectrogram image feature and subband-power-distribution image feature extract local statistics such as mean and variance from the spectrogram. We argue that such simple image statistics cannot well capture complex texture details of the spectrogram. Thus, we propose to extract pseudo-color CENTRIST features from the logarithm of Gammatone-like spectrogram. To well classify the sound event under the unknown noise condition, we propose a classifier-selection scheme, which automatically selects the most suitable classifier. The proposed approach is compared with the state of the art on the RWCP database, and demonstrates a superior performance.

Error Correcting Output Code (ECOC) has emerged as one of promising techniques for solving multi-class classification. In the ECOC framework, a multi-class problem is decomposed into several binary ones with a coding design scheme. Despite this, the suitable multi-class decomposition scheme is still ongoing research in machine learning. In this work, we propose a novel multi-class coding design method to mine the effective and compact class codes for multi-class classification. For a given n-class problem, this method decomposes the classes into subsets by embedding a structure of binary trees. We put forward a novel splitting criterion based on minimizing generalization errors across the classes. Then, a greedy search procedure is applied to explore the optimal tree structure for the problem domain. We run experiments on many multi-class UCI datasets. The experimental results show that our proposed method can achieve better classification performance than the common ECOC design methods.

This paper uses the keystroke dynamics in user authentication. The relationship between the distance metrics and the data template, for the first time, was analyzed and new distance based algorithm for keystroke dynamics classification was proposed. The results of the experiments on the CMU keystroke dynamics benchmark dataset¹ were evaluated with an equal error rate of 0.0614. The classifiers using the proposed distance metric outperform existing top performing keystroke dynamics classifiers which use traditional distance metrics.

Sentiment analysis has become a flourishing field of text mining and natural language processing. Sentiment analysis aims to determine whether the text is written to express positive, negative, or neutral emotions about a certain domain. Most sentiment analysis researchers focus on English texts, with very limited resources available for other complex languages, such as Arabic. In this study, the target was to develop an initial model that performs satisfactorily and measures Arabic Twitter sentiment by using machine learning approach, Naïve Bayes and Decision Tree for classification algorithms. The datasets used contains more than 2,000 Arabic tweets collected from Twitter. We performed several experiments to check the performance of the two algorithms classifiers using different combinations of text-processing functions. We found that available facilities for Arabic text processing need to be made from scratch or improved to develop accurate classifiers. The small functionalities developed by us in a Python language environment helped improve the results and proved that sentiment analysis in the Arabic domain needs lot of work on the lexicon side.

The accurate identification of encrypted data stream helps to regulate illegal data, detect network attacks and protect users' information. In this paper, a novel encrypted data stream identification algorithm is introduced. The proposed method is based on randomness characteristics of encrypted data stream. We use a l₁-norm regularized logistic regression to improve sparse representation of randomness features and Fuzzy Gaussian Mixture Model (FGMM) to improve identification accuracy. Experimental results demonstrate that the method can be adopted as an effective technique for encrypted data stream identification.

In this paper, we have investigated different methods to calibrate a 3-D scanning system consisting of multiple Kinect sensors. The main function of the scanning system is for the reconstruction of the full surface model of an object. In this work, we build a four-Kinect system that the Kinect range sensors are positioned around the target object. Each Kinect is responsible for capturing a small local model, and the local models found will be combined to become the full model. To build such a system, calibration of the poses among the Kinects is essential. We have tested a number of methods: using (1) a sphere, (2) a checker board and (3) a cube as the calibration object. After calibration, the results of method (1) and (2) are used in the multiple Kinect system for obtaining the 3-D model of a real object. Results are shown and compared. For method (3) we only performed the simulation test on finding the rotation between two Kinects and the result is promising. This is the first part of a long term project on building a full surface model capturing system. Such a system should be useful in robot vision, scientific research and many other industrial applications.

Named Entity Recognition (NER) plays an important role in many Natural Language Processing (NLP) applications such as; Information Extraction (IE), Question Answering (QA), Text Clustering, Text Summarization and Word Sense Disambiguation. This paper presents the development and implementation of domain independent system to recognize three types of Arabic named entities. The system works based on a set of domain independent grammar-rules along with Arabic part of speech tagger in addition to gazetteers and lists of trigger words. The experimental results shown, that the system performed as good as other systems with better results in some cases of cross-domains corpora.

To understand cyber citizens’ opinion accurately from Chinese news comments, the clear definition on nonsense is present, and a detection model based on logistic regression (LR) is proposed. The detection of nonsense can be treated as a binary-classification problem. Besides of traditional lexical features, we propose three kinds of features in terms of emotion, structure and relevance. By these features, we train an LR model and demonstrate its effect in understanding Chinese news comments. We find that each of proposed features can significantly promote the result. In our experiments, we achieve a prediction accuracy of 84.3% which improves the baseline 77.3% by 7%.

Getting stuck in local maxima is a problem that arises while learning Bayesian networks (BNs) structures. In this paper, we studied a recently proposed Markov chain Monte Carlo (MCMC) sampler, called the Neighbourhood sampler (NS), and examined how efficiently it can sample BNs when local maxima are present. We assume that a posterior distribution f(N,E|D) has been defined, where D represents data relevant to the inference, N and E are the sets of nodes and directed edges, respectively. We illustrate the new approach by sampling from such a distribution, and inferring BNs. The simulations conducted in this paper show that the new learning approach substantially avoids getting stuck in local modes of the distribution, and achieves a more rapid rate of convergence, compared to other common algorithms e.g. the MCMC Metropolis-Hastings sampler.

The positioning and navigation systems based on Global Positioning System (GPS) have been developed over past decades and have been widely used for outdoor environment. However, high-rise buildings or indoor environments can block the satellite signal. Therefore, many indoor positioning methods have been developed to respond to this issue. In addition to the distance measurements using sonar and laser sensors, this study aims to develop a method by integrating a monocular simultaneous localization and mapping (MonoSLAM) algorithm with an inertial measurement unit (IMU) to build an indoor positioning system. The MonoSLAM algorithm measures the distance (depth) between the image features and the camera. With the help of Extend Kalman Filter (EKF), MonoSLAM can provide real-time position, velocity and camera attitude in world frame. Since the feature points will not always appear and can't be trusted at any time, a wrong estimation of the features will cause the estimated position diverge. To overcome this problem, a multisensor fusion algorithm was applied in this study by using the multi-rate Kalman Filter. Finally, from the experiment results, the proposed system was verified to be able to improve the reliability and accuracy of the MonoSLAM by integrating the IMU measurements.

A lunar rover requires an accurate localisation system in order to operate in an uninhabited environment. However, every additional piece of equipment mounted on it drastically increases the overall cost of the mission. This paper reports a possible solution for a micro-rover using a sole monocular omnidirectional camera. Our approach relies on a combination of feature tracking and template matching for Visual Odometry. The results are afterwards refined using a Graph-Based SLAM algorithm, which also provides a sparse reconstruction of the terrain. We tested the algorithm on a lunar rover prototype in a lunar analogue environment and the experiments show that the estimated trajectory is accurate and the combination with the template matching algorithm allows an otherwise poor detection of spot turns.

This paper proposes a technology which enables healthy human brain to control electronic wheelchair movement. The method involves acquiring electroencephalograph (EEG) data from specific channels using Emotiv Software Development Kit (SDK) into Windows based application in a tablet PC to be preprocessed and classified. The aim of this research is to increase the accuracy rate of the brain control system by applying Support Vector Machine (SVM) as machine learning algorithm. EEG samples are taken from several respondents with disabilities but still have healthy brain to pick most suitable EEG channel which will be used as a proper learning input in order to simplify the computational complexity. The controller system based on Arduino microcontroller and combined with .NET based software to control the wheel movement. The result of this research is a brain-controlled electric wheelchair with enhanced and optimized EEG classification.