This PDF file contains the front matter associated with SPIE Proceedings Volume 10341, including the Title Page, Copyright information, Table of Contents, Introduction (if any), and Conference Committee listing.

Local feature detection is a fundamental module in several mobile vision applications such as mobile object recognition and mobile visual search. The effectiveness and the efficiency of a local feature detector decide to what extent it is suitable for a mobile application. Over the past decades, several local feature detectors have been developed. In this paper, we are interested in FAST (Features from Accelerated Segment Test) local feature detector for its efficiency. However, FAST detector shows poor robustness against both scale and rotation changes. Therefore, we aim at enhancing FAST robustness against both scale and rotation changes while maintaining good efficiency. To this end, we propose a Scalable and Oriented FAST-based local Feature detector (SOFF). A comprehensive comparison against FAST detector and its variants is performed on benchmark datasets. Experimental results demonstrate that SOFF detector outperforms other FAST-based detectors in many cases. Furthermore, it is efficient to compute, thereby suitable for mobile vision applications.

Recurrent Neural Networks (RNNs) are extensively used for time-series modeling and prediction. We propose an approach for automatic construction of a binary classifier based on Long Short-Term Memory RNNs (LSTM-RNNs) for detection of a vehicle passage through a checkpoint. As an input to the classifier we use multidimensional signals of various sensors that are installed on the checkpoint. Obtained results demonstrate that the previous approach to handcrafting a classifier, consisting of a set of deterministic rules, can be successfully replaced by an automatic RNN training on an appropriately labelled data.

This paper introduces a novel method of moving object classification in Infrared and Visible spectra. This method is based on a data-mining process by combining a set of best features based on shape, texture and motion. The proposed method relies either on visible spectrum or on infrared spectrum according to weather conditions (sunny days, rain, fog, snow, etc.) and timing of the video acquisition. Experimental studies are carried out to prove the efficiency of our predictive models to classify moving objects and the originality of our process with intelligent fusion of VIS-IR spectra.

Currently, there are several fall detection systems based on video analysis. However, these systems have not yet reached the desired level of appropriateness and robustness. To reduce the risk of falling in insecure environments, a new method is developed in this paper to detect and predict human fall detection. We adopt, in this approach, a Block Matching motion estimation algorithm based on acceleration and changes of the human body silhouette area, which are obtained from a single surveillance camera. It presents an algorithm to accelerate the fall detection system on based on a local adjustment of the velocity field.

Human behavior analysis in a video scene is based on motion feature extraction and recognition, many information is hidden behind gesture, sudden motion and walking, many research works tried to model and then recognize human behavior through motion analysis. In our work we will focus on motion extraction and tracking with some enhancements comparing to previous approaches. The first phase of behavior analysis is the extraction of regions in the video in motion. The second phase is devoted to motion tracking using Kalman filter. Performances evaluation are based on both visual quality and numerical values of MSE(Mean Squared Error).

In recent years, Computer vision has become a very active field. This field includes methods for processing, analyzing, and understanding images. The most challenging problems in computer vision are image classification and object recognition. This paper presents a new approach for object recognition task. This approach exploits the success of the Very Deep Convolutional Neural Network for object recognition. In fact, it improves the convolutional layers by adding recurrent connections. This proposed approach was evaluated on two object recognition benchmarks: Pascal VOC 2007 and CIFAR-10. The experimental results prove the efficiency of our method in comparison with the state of the art methods.

Grasping of unknown objects with neither appearance data nor object models given in advance is very important for robots that work in an unfamiliar environment. The goal of this paper is to quickly synthesize an executable grasp for one unknown object by using cylinder searching on a single point cloud. Specifically, a 3D camera is first used to obtain a partial point cloud of the target unknown object. An original method is then employed to do post treatment on the partial point cloud to minimize the uncertainty which may lead to grasp failure. In order to accelerate the grasp searching, surface normal of the target object is then used to constrain the synthetization of the cylinder grasp candidates. Operability analysis is then used to select out all executable grasp candidates followed by force balance optimization to choose the most reliable grasp as the final grasp execution. In order to verify the effectiveness of our algorithm, Simulations on a Universal Robot arm UR5 and an under-actuated Lacquey Fetch gripper are used to examine the performance of this algorithm, and successful results are obtained.

The aim of this paper is to present the novel proposition of the human motion modelling and recognition approach that enables real time MoCap signal evaluation. By motions (actions) recognition we mean classification. The role of this approach is to propose the syntactic description procedure that can be easily understood, learnt and used in various motion modelling and recognition tasks in all MoCap systems no matter if they are vision or wearable sensor based. To do so we have prepared extension of Gesture Description Language (GDL) methodology that enables movements description and real-time recognition so that it can use not only positional coordinates of body joints but virtually any type of discreetly measured output MoCap signals like accelerometer, magnetometer or gyroscope. We have also prepared and evaluated the cross-platform implementation of this approach using Lua scripting language and JAVA technology. This implementation is called Data Driven GDL (DD-GDL). In tested scenarios the average execution speed is above 100 frames per second which is an acquisition time of many popular MoCap solutions.

Speaker gender identification is considered among the most important tools in several multimedia applications namely in automatic speech recognition, interactive voice response systems and audio browsing systems. Gender identification systems performance is closely linked to the selected feature set and the employed classification model. Typical techniques are based on selecting the best performing classification method or searching optimum tuning of one classifier parameters through experimentation. In this paper, we consider a relevant and rich set of features involving pitch, MFCCs as well as other temporal and frequency-domain descriptors. Five classification models including decision tree, discriminant analysis, nave Bayes, support vector machine and k-nearest neighbor was experimented. The three best perming classifiers among the five ones will contribute by majority voting between their scores. Experimentations were performed on three different datasets spoken in three languages: English, German and Arabic in order to validate language independency of the proposed scheme. Results confirm that the presented system has reached a satisfying accuracy rate and promising classification performance thanks to the discriminating abilities and diversity of the used features combined with mid-level statistics.

To control the state of horses in the born, breeders needs a monitoring system with a surveillance camera that can identify and distinguish between horses. We proposed in [5] a method of horse’s identification at a distance using the frontal facial biometric modality. Due to the change of views, the face recognition becomes more difficult. In this paper, the number of images used in our THoDBRL’2015 database (Tunisian Horses DataBase of Regim Lab) is augmented by adding other images of other views. Thus, we used front, right and left profile face’s view. Moreover, we suggested an approach for multiview face recognition. First, we proposed to use the Gabor filter for face characterization. Next, due to the augmentation of the number of images, and the large number of Gabor features, we proposed to test the Deep Neural Network with the auto-encoder to obtain the more pertinent features and to reduce the size of features vector. Finally, we performed the proposed approach on our THoDBRL’2015 database and we used the linear SVM for classification.

In our daily activities we perform prediction or anticipation when interacting with other humans or with objects. Prediction of human activity made by computers has several potential applications: surveillance systems, human computer interfaces, sports video analysis, human-robot-collaboration, games and health-care. We propose a system capable of recognizing and predicting human actions using supervised classifiers trained with automatically labeled data evaluated in our human activity RGB-D dataset (recorded with a Kinect sensor) and using only the position of the main skeleton joints to extract features. Using conditional random fields (CRFs) to model the sequential nature of actions in a sequence has been used before, but where other approaches try to predict an outcome or anticipate ahead in time (seconds), we try to predict what will be the next action of a subject. Our results show an activity prediction accuracy of 89.9% using an automatically labeled dataset.

Two systems of eyes and hand gestures recognition are used to control mobile devices. Based on a real-time video streaming captured from the device's camera, the first system recognizes the motion of user's eyes and the second one detects the static hand gestures. To avoid any confusion between natural and intentional movements we developed a system to fuse the decision coming from eyes and hands gesture recognition systems. The phase of fusion was based on decision tree approach. We conducted a study on 5 volunteers and the results that our system is robust and competitive.

Security and surveillance are vital issues in today’s world. The recent acts of terrorism have highlighted the urgent need for efficient surveillance. There is indeed a need for an automated system for video surveillance which can detect identity and activity of person. In this article, we propose a new paradigm to recognize an aggressive human behavior such as boxing action. Our proposed system for human activity detection includes the use of a fusion between Spatio Temporal Interest Point (STIP) and Histogram of Oriented Gradient (HoG) features. The novel feature called Spatio Temporal Histogram Oriented Gradient (STHOG). To evaluate the robustness of our proposed paradigm with a local application of HoG technique on STIP points, we made experiments on KTH human action dataset based on Multi Class Support Vector Machines classification. The proposed scheme outperforms basic descriptors like HoG and STIP to achieve 82.26% us an accuracy value of classification rate.

The paper handles the problem of segmentation of handwriting on mobile devices. Many applications have been developed in order to facilitate the recognition of handwriting and to skip the limited numbers of keys in keyboards and try to introduce a space of drawing for writing instead of using keyboards. In this one, we will present a mobile theory for the segmentation of for handwriting uses PerTOHS theory, Perceptual Theory of On line Handwriting Segmentation, where handwriting is defined as a sequence of elementary and perceptual codes. In fact, the theory analyzes the written script and tries to learn the handwriting visual codes features in order to generate new ones via the generated perceptual sequences. To get this classification we try to apply the Beta-elliptic model, fuzzy detector and also genetic algorithms in order to get the EPCs (Elementary Perceptual Codes) and GPCs (Global Perceptual Codes) that composed the script. So, we will present our Android application M-PerTOHS for segmentation of handwriting.

This paper highlights a new method for online Arabic handwriting recognition based on graphemes segmentation. The main contribution of our work is to explore the utility of Beta-elliptic model in segmentation and features extraction for online handwriting recognition. Indeed, our method consists in decomposing the input signal into continuous part called graphemes based on Beta-Elliptical model, and classify them according to their position in the pseudo-word. The segmented graphemes are then described by the combination of geometric features and trajectory shape modeling. The efficiency of the considered features has been evaluated using feed forward neural network classifier. Experimental results using the benchmarking ADAB Database show the performance of the proposed method.

Tracking moving objects is an area increasingly known in computer vision field. It plays a very important role in human-computer interaction. In this context we have developed a hand tracking and gesture recognition system that allows interaction with the machine in an intuitive and natural way. To ensure the tracking we apply the Kalman filter and detect the optimal points of the hand in order to determine the gesture expressed by user.

Using camera traps in animal ecology studies has increased because it facilitates the work of biologists and allows them to obtain information that otherwise would be impossible. A large number of photographs are capturing with this wildlife photography technique making difficult their posterior analysis. This paper presents a method to automatically identify the images with at least one animal and to classify them between birds and mammals. In this work a fuzzy classifier and a matched filter were used to identify the image with animals and to segment the images. An artificial neural network was employed to classify the segments between birds and mammals. We obtained a classification accuracy of 73.1% validating the model over real camera trap sessions. The database includes several difficulties, as the constant changes in the scene by climatic factors or animals partially occluded by the environment. This method was implemented in a software that is currently using in the Alexander von Humboldt Biological Resources Research Institute for studies of biodiversity in Colombia.

This paper considers the issue of fully automatic emotion classification on 2D faces. In spite of the great effort done in recent years, traditional machine learning approaches based on hand-crafted feature extraction followed by the classification stage failed to develop a real-time automatic facial expression recognition system. The proposed architecture uses Convolutional Neural Networks (CNN), which are built as a collection of interconnected processing elements to simulate the brain of human beings. The basic idea of CNNs is to learn a hierarchical representation of the input data, which results in a better classification performance. In this work we present a block-based CNN algorithm, which uses noise, as data augmentation technique, and builds batches with a balanced number of samples per class. The proposed architecture is a very simple yet powerful CNN, which can yield state-of-the-art accuracy on the very competitive benchmark algorithm of the Extended Cohn Kanade database.

Pointer meters are frequently applied to industrial production for they are directly readable. They should be calibrated regularly to ensure the precision of the readings. Currently the method of manual calibration is most frequently adopted to accomplish the verification of the pointer meter, and professional skills and subjective judgment may lead to big measurement errors and poor reliability and low efficiency, etc. In the past decades, with the development of computer technology, the skills of machine vision and digital image processing have been applied to recognize the reading of the dial instrument. In terms of the existing recognition methods, all the parameters of dial instruments are supposed to be the same, which is not the case in practice. In this work, recognition of pointer meter reading is regarded as an issue of pattern recognition. We obtain the features of a small area around the detected point, make those features as a pattern, divide those certified images based on Gradient Pyramid Algorithm, train a classifier with the support vector machine (SVM) and complete the pattern matching of the divided mages. Then we get the reading of the pointer meter precisely under the theory of dynamic three points make a line (DTPML), which eliminates the error caused by tiny differences of the panels. Eventually, the result of the experiment proves that the proposed method in this work is superior to state-of-the-art works.

Recognizing the online Arabic handwritten script has been gaining more interest because of the impressive advances in mobile device requiring more and more intelligent handwritten recognizers. Since it was demonstrated within many previous research that Deep Neural Networks (DNN) exhibit a great performance, we propose in this work a new system based on a DNN in which we try to optimize the training process by a smooth construct of the deep architecture. The Output’s error of each unit in the previous layer will be computed and only the smallest error will be maintained in the next iteration. This paper uses LMCA database for training and testing data. The experimental study reveals that our proposed DBNN using generated Bottleneck features can outperform state of the art online recognizers.

In this paper, a scalable parallel image recognition method, CS-BoW (Class-Specific Bag of Words), is proposed. CSBoW builds submodels for each class using weighted BoW in training phase. In test phase, CS-BoW calculates and sorts the coding residual for every class of each test image, then each test image is assigned to the class which has the smallest coding residual. The proposed CS-BoW is easy to be scaled, the only thing to do to extend training dataset with a new class is to build a submodel for the new class. No extra calculation for existing data. CS-BoW calculates coding residual of each test image for every class, so it is convenient to obtain Top N accuracy, while many other image recognition methods can only release Top 1 accuracy. The experimental results show that CS-BoW achieves comparable accuracy in very short time if it is run parallelly, it can be as fast as 0.05s per image in caltech 101 and caltech 256.

The motivation for this paper is to initially propose and evaluate two new kinematics models that were developed to describe motion capture (MoCap) data of karate techniques. We decided to develop this novel proposition to create the model that is capable to handle actions description both from multimedia and professional MoCap hardware. For the evaluation purpose we have used 25-joints data with karate techniques recordings acquired with Kinect version 2. It is consisted of MoCap recordings of two professional sport (black belt) instructors and masters of Oyama Karate. We have selected following actions for initial analysis: left-handed furi-uchi punch, right leg hiza-geri kick, right leg yoko-geri kick and left-handed jodan-uke block. Basing on evaluation we made we can conclude that both proposed kinematics models seems to be convenient method for karate actions description. From two proposed variables models it seems that global might be more useful for further usage. We think that because in case of considered punches variables seems to be less correlated and they might also be easier to interpret because of single reference coordinate system. Also principal components analysis proved to be reliable way to examine the quality of kinematics models and with the plot of the variable in principal components space we can nicely present the dependences between variables.

In this paper we consider a task of improving optical character recognition (OCR) results of document fields on low-quality and average-quality images using N-gram models. Cyrillic fields of Russian Federation internal passport are analyzed as an example. Two approaches are presented: the first one is based on hypothesis of dependence of a symbol from two adjacent symbols and the second is based on calculation of marginal distributions and Bayesian networks computation. A comparison of the algorithms and experimental results within a real document OCR system are presented, it's showed that the document field OCR accuracy can be improved by more than 6% for low-quality images.

In this paper, we introduce slant detection method based on Fast Hough Transform calculation and demonstrate its application in industrial system for Russian passports recognition. About 1.5% of this kind of documents appear to be slant or italic. This fact reduces recognition rate, because Optical Recognition Systems are normally designed to process normal fonts. Our method uses Fast Hough Transform to analyse vertical strokes of characters extracted with the help of x-derivative of a text line image. To improve the quality of detector we also introduce field grouping rules. The resulting algorithm allowed to reach high detection quality. Almost all errors of considered approach happen on passports of nonstandard fonts, while slant detector works in appropriate way.

This paper explores method of layer-by-layer training for neural networks to train neural network, that use approximate calculations and/or low precision data types. Proposed method allows to improve recognition accuracy using standard training algorithms and tools. At the same time, it allows to speed up neural network calculations using fast-processed approximate calculations and compact data types. We consider 8-bit fixed-point arithmetic as the example of such approximation for image recognition problems. In the end, we show significant accuracy increase for considered approximation along with processing speedup.

Brain computer interface (BCI) systems based on the steady state visual evoked potential (SSVEP) provide higher information transfer rates and require shorter training time than BCI systems using other brain signals. It has been widely used in brain science, rehabilitation engineering, biomedical engineering and intelligent information processing. In this paper, we present a real-time mobile phone dialing system based on SSVEP, and it is more portable than other dialing system because the flashing dial interface is set on a small tablet. With this online BCI system, we can take advantage of this system based on SSVEP to identify the specific frequency on behalf of a number using canonical correlation analysis (CCA) method and dialed out successfully without using any physical movements such as finger tapping. This phone dialing system will be promising to help disable patients to improve the quality of lives.

Many works have devoted to exploring local region information including both the information of the local features in local region and their spatial relationships, but none of these can provide a compact representation of the information. To achieve this, we propose a new approach named Local Visual Similarity (LVS). LVS first calculates the similarities among the local features in a local region and then forms these similarities as a single vector named LVS descriptor. In our experiments, we show that LVS descriptor can preserve local region information with low dimensionality. Besides, experimental results on two public datasets also demonstrate the effectiveness of LVS descriptor.

Most of the existing solutions predicting visual attention focus solely on referenced 2D images and disregard any depth information. This aspect has always represented a weak point since the depth is an inseparable part of the biological vision. This paper presents a novel method of saliency map generation based on results of our experiments with egocentric visual attention and investigation of its correlation with perceived depth. We propose a model to predict the attention using superpixel representation with an assumption that contrast objects are usually salient and have a sparser spatial distribution of superpixels than their background. To incorporate depth information into this model, we propose three different depth techniques. The evaluation is done on our new RGB-D dataset created by SMI eye-tracker glasses and KinectV2 device.

In this paper, considering the atmospheric refractive index, we present an approach to extract the recording date and geo-location from videos, based on the alteration of the shadow length. The paper carefully takes different information (photographed date, the length of the selected object) of the given video into consideration and forms a comprehensive approach to extract the temporal and spatial information of the given video. On the basis of this approach, we analyze the shadow length data of a chosen object from a real video and extract the temporal and spatial information of the video. Compared with the actual information, the error is less than 1%, which proves the validity of our approach.

Reducing noise in a video sequence is of vital important in many real-world applications. One popular method is block matching collaborative filtering. However, the main drawback of this method is that noise standard deviation for the whole video sequence is known in advance. In this paper, we present a tensor based denoising framework that considers 3D patches instead of 2D patches. By collecting the similar 3D patches non-locally, we employ the low-rank tensor decomposition for collaborative filtering. Since we specify the non-informative prior over the noise precision parameter, the noise variance can be inferred automatically from observed video data. Therefore, our method is more practical, which does not require knowing the noise variance. The experimental on video denoising demonstrates the effectiveness of our proposed method.

The work in this paper is focused around visual ADAS (Advanced Driver Assistance Systems). We introduce forward rectification – a technique for making computer vision algorithms more robust against camera mount point and mount angles. Using the technique can increase the quality of recognition as well as lower the dimensionality for algorithm invariance, making it possible to apply simpler affine-invariant algorithms for applications that require projective invariance. Providing useful results this rectification requires thorough calibration of the camera, which can be done automatically or semi-automatically. The technique is of general nature and can be applied to different algorithms, such as pattern matching detectors, convolutional neural networks. The applicability of the technique is demonstrated on HOG-based car detector detection rate.

The procedure of colorimetric calibration of the camera ensures accurate and repeatable acquisition of the scene colors. The most common approach defines the calibration only as the color transformation between the camera image colors and colorimetric colors. The only condition for image acquisition is an uniform illumination of the scene. Unfortunately, such assumption does not include many distortions caused by image acquisition. One of them is a vignetting, which can be described as a decrease of the light intensity from the image center to the image corners. This phenomenon causes the same effects as non-uniform illumination, which is the change of color values of the same object depending on image coordinates. This paper is an attempt to analyze the influence of vignetting correction on the results of the camera colorimetric calibration. The conducted experiment in uniform light conditions shows that the improvement of calibration quality depends on the chosen vignetting correction method.

In this work we describe an approach to real-time image search in large databases robust to variety of query distortions such as lighting alterations, projective distortions or digital noise. The approach is based on the extraction of keypoints and their descriptors, random hierarchical clustering trees for preliminary search and RANSAC for refining search and result scoring. The algorithm is implemented in Snapscreen system which allows determining a TV-channel and a TV-show from a picture acquired with mobile device. The implementation is enhanced using preceding localization of screen region. Results for the real-world data with different modifications of the system are presented.

Ensemble learning has succeeded in the growth of stability and clustering accuracy, but their runtime prohibits them from scaling up to real-world applications. This study deals the problem of selecting a subset of the most pertinent features for every cluster from a dataset. The proposed method is another extension of the Random Forests approach using self-organizing maps (SOM) variants to unlabeled data that estimates the out-of-bag feature importance from a set of partitions. Every partition is created using a various bootstrap sample and a random subset of the features. Then, we show that the process internal estimates are used to measure variable pertinence in Random Forests are also applicable to feature selection in unsupervised learning. This approach aims to the dimensionality reduction, visualization and cluster characterization at the same time. Hence, we provide empirical results on nineteen benchmark data sets indicating that RFS can lead to significant improvement in terms of clustering accuracy, over several state-of-the-art unsupervised methods, with a very limited subset of features. The approach proves promise to treat with very broad domains.

The adaptive nonlinear predictor is proposed for digital image compression method based on differential pulse code modulation. Greham predictor is parameterized for this. Proposed predictor works in different ways depending on the local image contours. A special feature is offered for estimation the contour direction and intensity in the neighborhood of the current pixel. The parameters of the proposed predictor are calculated by rapid training procedure before the actual compression. This procedure minimizes the sum of absolute values of prediction errors. Theoretical computational complexity of the proposed predictor is shown. Considered predictors are compared in real images by computational experiments. The win of proposed algorithm is demonstrated. In addition, the gain of compression method based on differential pulse code modulation with the proposed predictor against JPEG compression method is demonstrated.

Text detection in natural scenes holds great importance in the field of research and still remains a challenge and an important task because of size, various fonts, line orientation, different illumination conditions, weak characters and complex backgrounds in image. The contribution of our proposed method is to filtering out complex backgrounds by combining three strategies. These are enhancing the edge candidate detection in HSV space color, then using MSER candidate detection to get different masks applied in HSV space color as well as gray color. After that, we opt for the Stroke Width Transform (SWT) and heuristic filtering. Such strategies are followed so as to maximize the capacity of zones text pixels candidates and distinguish between text boxes and the rest of the image. The non-text components are filtered by classifying the characters candidates based on Support Vector Machines (SVM) using Histogram of Oriented Gradients (HOG) features. Finally we apply boundary box localization after a stage of word grouping where false positives are eliminated by geometrical properties of text blocks. The proposed method has been evaluated on ICDAR 2013 scene text detection competition dataset and the encouraging experiments results demonstrate the robustness of our method.

Triangular surface are now widely used for modeling three-dimensional object, since these models are very high resolution and the geometry of the mesh is often very dense, it is then necessary to remesh this object to reduce their complexity, the mesh quality (connectivity regularity) must be ameliorated. In this paper, we review the main methods of semi-regular remeshing of the state of the art, given the semi-regular remeshing is mainly relevant for wavelet-based compression, then we present our method for re-meshing based trust region spherical geometry image to have good scheme of 3d mesh compression used to deform 3D meh based on Multi library Wavelet Neural Network structure (MLWNN). Experimental results show that the progressive re-meshing algorithm capable of obtaining more compact representations and semi-regular objects and yield an efficient compression capabilities with minimal set of features used to have good 3D deformation scheme.

An iterative algorithm is proposed for blind multi-image deblurring of binary images. The binarity is the only prior restriction imposed on the image. Image formation model assumes convolution with arbitrary kernel and addition of a constant value. Penalty functional is composed using binarity constraint for regularization. The algorithm estimates the original image and distortion parameters by alternate reduction of two parts of this functional. Experimental results for natural (non-synthetic) data are present.

Automatic identification of television programs in the TV stream is an important task for operating archives. This article proposes a new spatio-temporal approach to identify the programs in TV stream into two main steps: First, a reference catalogue for video features visual jingles built. We operate the features that characterize the instances of the same program type to identify the different types of programs in the flow of television. The role of video features is to represent the visual invariants for each visual jingle using appropriate automatic descriptors for each television program. On the other hand, programs in television streams are identified by examining the similarity of the video signal for visual grammars in the catalogue. The main idea of the identification process is to compare the visual similarity of the video signal features in the flow of television to the catalogue. After presenting the proposed approach, the paper overviews encouraging experimental results on several streams extracted from different channels and compounds of several programs.

Frame difference method is a good method for motion segmentation, but its result contains much wrong motion regions and incomplete motion objects. In this paper we combine variational method with frame difference method to propose two motion segmentation models, and the proposed models are based on different invariance assumptions. The models can detect motion objects and make up for the inadequacy of frame differential method with smooth terms. Experimental results show that the proposed models can detect motion objects better.

It is common that textures occur in real-word color image, moreover, textures could cause difficulties in image segmentation. For the purpose of solving those difficulties, we put forward a new model. In this model we only need the structural and oscillating components’ information of the real color image. This model is based on the VO model, MTV and active contour models. We will use the fast Split Bregman algorithm to solve this model. The results of our model is mentioned in numerical experiments.

In the field of image segmentation, discriminative models have shown promising performance. Generally, every such model begins with the extraction of numerous features from annotated images. Most authors create their discriminative model by using many features without using any selection criteria. A more reliable model can be built by using a framework that selects the important variables, from the point of view of the classification, and eliminates the unimportant once. In this article we present a framework for feature selection and data dimensionality reduction. The methodology is built around the random forest (RF) algorithm and its variable importance evaluation. In order to deal with datasets so large as to be practically unmanageable, we propose an algorithm based on RF that reduces the dimension of the database by eliminating irrelevant features. Furthermore, this framework is applied to optimize our discriminative model for brain tumor segmentation.

Image segmentation is often used in medical image processing. This crucial task can affect all results obtained from the further steps of image analysis. In nuclear medicine emission tomography imaging, where acquired and reconstructed images contain large noise and high blurring level, segmentation and tumour boundaries delineation can be very challenging task. Many from already existing image segmentation methods are based on clustering. In this work, we have tested and implemented a few clustering based methods. We have mainly focused on k-means related algorithms to evaluate and compare their accuracy. In this group we have chosen k-means algorithm, k-medoid clustering, and fuzzy C-means (FCM) method. Results for all methods were verified using the gold standard obtained from anatomical image co-registered and emission tomography dataset. Numerical values of both datasets matching were calculated using the Jaccard index. Results were compared with standard segmentation algorithm based on fixed threshold (standardized uptake value - SUV with threshold 2.5), which is a commonly used standard in clinical practice and also with previously implemented and verified methods (including game theoretical algorithm). For all tested methods we have obtained very similar results, comparable to SUV 2.5 threshold method but worse than the game theoretic method.

Watershed transform is widely used in image segmentation. In literature, this transform is computed by various algorithms among which the M-border kernel algorithm [1]. This algorithm computes the watershed transform in the framework of edge weighted graphs. It is based on a local property that makes it adapted to parallelization. In this paper we propose a parallel implementation of this algorithm. We start by studying the data dependencies problematic that it raises. We give then an approach that allows overcoming this problematic based on an alternated edges processing strategy. The implementation of this strategy on a shared memory multicore architecture using a Single Program Multiple Data (SPMD) approach proves its effectiveness. In fact, experimental results show that our implementation achieves a relative speedup factor of 2.8 using 4 processors over the performance of the sequential algorithm using a single processor on the same system.

In this paper a new attack aimed at DM-QIM watermark detection and extraction is proposed, based on known histogram-based attacks against non-dithered QIM algorithm. Key recovery results obtained during proposed attack shows that for a sets of watermarked images it is possible to outperform known watermark detection attacks proposed for DM-QIM in case if non-local image statistics is utilized. Also several known techniques for secure key synchronization are recommended to improve robustness of existing DM-QIM robustness against proposed attack.

The presence of errors in tomographic image may lead to misdiagnosis when computed tomography (CT) is used in medicine, or the wrong decision about parameters of technological processes when CT is used in the industrial applications. Two main reasons produce these errors. First, the errors occur on the step corresponding to the measurement, e.g. incorrect calibration and estimation of geometric parameters of the set-up. The second reason is the nature of the tomography reconstruction step. At the stage a mathematical model to calculate the projection data is created. Applied optimization and regularization methods along with their numerical implementations of the method chosen have their own specific errors. Nowadays, a lot of research teams try to analyze these errors and construct the relations between error sources. In this paper, we do not analyze the nature of the final error, but present a new approach for the calculation of its distribution in the reconstructed volume. We hope that the visualization of the error distribution will allow experts to clarify the medical report impression or expert summary given by them after analyzing of CT results. To illustrate the efficiency of the proposed approach we present both the simulation and real data processing results.

In this paper, we propose an expansion of convolutional neural network (CNN) input features based on Hough Transform. We perform morphological contrasting of source image followed by Hough Transform, and then use it as input for some convolutional filters. Thus, CNNs computational complexity and the number of units are not affected. Morphological contrasting and Hough Transform are the only additional computational expenses of introduced CNN input features expansion. Proposed approach was demonstrated on the example of CNN with very simple structure. We considered two image recognition problems, that were object classification on CIFAR-10 and printed character recognition on private dataset with symbols taken from Russian passports. Our approach allowed to reach noticeable accuracy improvement without taking much computational effort, which can be extremely important in industrial recognition systems or difficult problems utilising CNNs, like pressure ridge analysis and classification.

The paper considers the problem of fast nonlinear mapping of hyperspectral data. The analysis of various ways to speed-up the nonlinear mapping algorithm is performed, including stochastic algorithms, methods based on space partitioning, interpolation techniques, and parallel implementations using modern parallel architecture. The general scheme for hyperspectral data processing that summarizes the analyzed methods and algorithms is given with recommendations. Experimental results for the proposed technique are presented for well-known hyperspectral images. Possible applications of the technique for hyperspectral image analysis are discussed in the paper.

A new method for endoscopic image analysis in real time is presented. This method allows improving accuracy and helps avoiding subjectivity when performing endoscopic image classification. The method is based on use of Scale– invariant feature transform detector and computation of gastric mucosa pit–pattern skeletons. Subsequent use of the “Bag of visual words” method (“Bag of features”) and K–means method for key points clustering allows image classification with more than 85% accuracy.

Algorithms for detection of informative image fragments are proposed. Algorithms detect fragments of two types. Fragments of the first type («homogeneous fragments») are used to estimate the noise in the image. Fragments of the second type («stepped fragment») are used to estimate the frequency response of distortion system. Thus, these fragments can be used for assessment of image distortions and image quality in natural scenarios.

Color quantization is an important operation in the field of color image processing. In this paper, we consider a usefulness of the new DSCSI metric to assessment of quantized images. This metric is shown in the background of other useful image quality metrics to evaluate the color image differences and it has also been proven that DSCSI metric achieves the highest correlation coefficients with MOS. For further veriffcation DSCSI metric the combined methods that use to initialize the results of well-known splitting algorithms such as POP, MC, Wu etc. were tested. Experimental results of such combined methods indicate that the Wu+KM approach leading to the best quantized images in the sense of DSCSI metric.

Color image quantization is an often used in such tasks as image compression and image segmentation. In the paper, we continue to consider the usefulness of the new DSCSI metric for evaluating quantized images. Our use of the DSCSI metric confirmed that the color quantization in the CIELAB color space is better than in the basic RGB color space. On several examples we found very good DSCSI suitability in the case of quantization with dithering. During the tests of different dithering algorithms the best results, in terms of DSCSI metric, reached the classical Floyd-Steinberg algorithm at error propagation level of 75-85%.

In this work, a new ensemble method for the task of category recognition in different environments is presented. The focus is on service robotic perception in an open environment, where the robot’s task is to recognize previously unseen objects of predefined categories, based on training on a public dataset. We propose an ensemble learning approach to be able to flexibly combine complementary sources of information (different state-of-the-art descriptors computed on color and depth images), based on a Markov Random Field (MRF). By exploiting its specific characteristics, the MRF ensemble method can also be executed as a Dynamic Classifier Selection (DCS) system. In the experiments, the committee- and topology-dependent performance boost of our ensemble is shown. Despite reduced computational costs and using less information, our strategy performs on the same level as common ensemble approaches. Finally, the impact of large differences between datasets is analyzed.

Mechatronics is a new methodology used to achieve an optimal design of an electromechanical product. This methodology is collection of practices, procedures and rules used by those who work in particular branch of knowledge or discipline. Education in mechatronics at the Polytechnic University Timisoara is organized on three levels: bachelor, master and PhD studies. These activities refer and to design the mechatronics systems. In this context the design, implementation and experimental study of a family of mechatronic demonstrator occupy an important place. In this paper, a variant for a mechatronic demonstrator based on the combination of the electrical and mechanical components is proposed. The demonstrator, named humanoid robot, is equivalent with an inverted pendulum. Is presented the analyze of components for associated functions of the humanoid robot. This type of development the mechatronic systems by the combination of hardware and software, offers the opportunity to build the optimal solutions.

This paper presents a comparison of four most recent ROS-based monocular SLAM-related methods: ORB-SLAM, REMODE, LSD-SLAM, and DPPTAM, and analyzes their feasibility for a mobile robot application in indoor environment. We tested these methods using video data that was recorded from a conventional wide-angle full HD webcam with a rolling shutter. The camera was mounted on a human-operated prototype of an unmanned ground vehicle, which followed a closed-loop trajectory. Both feature-based methods (ORB-SLAM, REMODE) and direct SLAMrelated algorithms (LSD-SLAM, DPPTAM) demonstrated reasonably good results in detection of volumetric objects, corners, obstacles and other local features. However, we met difficulties with recovering typical for offices homogeneously colored walls, since all of these methods created empty spaces in a reconstructed sparse 3D scene. This may cause collisions of an autonomously guided robot with unfeatured walls and thus limits applicability of maps, which are obtained by the considered monocular SLAM-related methods for indoor robot navigation.