Search

Secure routing is a difficult task because of the limited nature of wireless sensor network resources. This paper provides a Hidden Markov Model solution to identify malicious nodes in wireless sensor networks through prevention of black hole attack. Our proposed approach based on a new routing algorithm which analyses shortest path in order to avoid malicious node path. Our results demonstrate the success and the efficiency of our proposed routing algorithm.

Representation of knowledge remains a problem for models of machine learning (ML). The Paragraph vector is one of the current methods for embedding the text, where many parameters govern the utility of representation. In this context, we are addressing the effect, on the text classification area, of Paragraph Vector-Distributed Memory (PV-DM) as variant of doc2vec. In comparison, we apply other classification systems focused on doc2vec forms, and a collection of classifiers with current practices in this article. Then, we incorporate hybrid ML methods to improve the quality of classification. The experiments, on benchmarking dataset, prove that the results obtained are excellent, with 99% accuracy in the system based on the PV-DM with average method, and majority voting as a classifier.

In the last decade, the speaker recognition systems know a wide interest from many researchers of speech and speaker recognition technologies that led this technology to become a practical identity authentication means. This paper reviews the speaker recognition systems evolution from the first implementation to current trends. In this paper, we present a brief overview of the speaker recognition area by describing its modeling methodologies and different feature extraction techniques with several languages. Then, we will display some speaker recognition strengths and weaknesses and performance indications. Finally, we will classify different research on Arabic and Amazigh speaker identification with various extraction methods and modeling approaches that can potentially be investigated for future works.

With the development and emergence of the concept of the Web, as well as the advancement of social networks, forums, blogs and E-commerce a growing number of Arabs express their views and opinions using their native language on those platforms where we find a wealth of information that could be exploited. This information, whose size is expanding and which describes opinions, has become the subject of several research fields particularly: “Sentiment Analysis” or “Opinion Mining”. Sentiment Analysis considered a subfield of natural language processing (NLP), information retrieval and text mining. Currently, sentiment analysis is commonly used for English, while it is in its infancy in the Arabic language. The most important and critical step in feeling analysis is to select an appropriate technique for classifying feelings. In this paper, we will be reviewing the published work on the application of Deep Learning (DL) techniques to the problems of sentiment analysis. We focus on the Arabic language because of its importance, the large number of its speakers, the scarcity of Arabic resources on it and the challenges related to its work. We explain, categorize, summarize and compare the techniques offered in this area.

Image mosaicking is a combination of multiple algorithms that use a series of images or a video to obtain as a result a single image with a larger field of view. However, in some cases the input data contains dynamic elements that change their position from a frame to another and this results in the appearance of errors like ghosting and parallax effects. In this paper, we will present a method to obtain a quantitative measure for mosaics with dynamic elements and how to determine the areas that create these errors. In fact, this was done by using dense optical flow to estimate the motion of all pixels and by comparing the Structure Similarity Index Measure of these areas obtained between the result mosaic and the input images. The obtained quantitative measures show better results and reflect better the obtained visual results.

In last years, several biometric modalities are coming back into the field of education. Especially, face recognition modality that allows authenticating humans based on their facial features. The main objective of this paper is to implement the recognition system part for a global software that is intended to be used in many applications related to the educational field. Therefore, we propose to combine the face description method based on Local Gradient Probabilistic Pattern (LGPP), the two dimensional subspace methods, and machine learning classifiers. Firstly, we extract principal face component using the LGPP descriptor. Then, 2DDWT, 2DPCA, and 2DLDA subspace methods were implemented to reduce face features. Finally, support vector machine (SVM) and artificial neural network (ANN) based machine learning algorithms are applied to classify the set of the final features vectors. The experimental results on relevant databases demonstrate the effectiveness of the proposed system.

The normality tests are widely used to estimate the difference between an empirical distribution and the normal distribution. In this paper, we propose to apply the Jarque–Bera normality test (JB-test) in the face recognition field. In order to find out the face areas, which can be approximately represented using a normal distribution, we propose to build a new descriptor based on the JB- test and Local Binary Pattern (LBP) descriptor. To predict whether a pixel belongs to the homogeneous area, Jarque–Bera Local Binary Pattern (JB-LBP) computes the confidence interval basing on the JB- test and uses it later to generate a new pixel grayscale value. The obtained results show that the homogeneous areas pixels are in same way encoded. Therefore, we can easily distinguish them from the peaks areas. Moreover, to reveal that the proposed face representation gives better performances, we combine it with the K-means clustering, the K-Nearest Neighbors (KNN) and the Multi-layers Perceptron (MLP).

Interactions between trophic and halieutic populations, have often been ignored in most mathematical models that are interested in prey–predator dynamics. However, the presence of vegetation in hydraulic environments, is important, and it has a direct impact on the life cycle of many fishes. In fact, aquatic populations are essential for providing oxygen, food and also shelter for grazers. For this, we suggest a mathematical and optimization approach, in an attempt to discuss the possibility of finding effective harvesting control strategies that aim to optimize the fishing efforts without affecting the trophic-halieutic populations and compromising with the interests of fishermen. The model is in the form of three ordinary differential equations, and which describes dynamics of multi-species of a fishery that includes grazer–predator fishes and aquatic plants, during fishing periods. We study the stability of the proposed differential system, and we suggest harvesting optimal control approaches for the environmental sustainability and bioeconomic cases after the introduction of two control functions in the model and which represent the efforts of fishing in grazer and predator populations respectively. The two optimal controls are characterized in the first case, based on Pontryagin’s maximum principle, and we seek their analytical formulations in the second case, when they are singular functions. Hence, these two optimal control approaches lead to two two-point boundary value problems which are resolve numerically based on the forward-backward sweep method with the incorporation of iterative Runge-Kutta fourth order progressive-regressive schemes.

In this paper, we studied image segmentation to which we applied a combination of the genetic algorithm and the cooperation between unsupervised classification by K-means and contour detection by the Sobel filter to improve image segmentation results by the K-means method alone. First, we will apply the segmentation process by combining two methods in the following way: We have hybridized the K-means method which is used to classify pixels into classes (regions), with the Sobel gradient filter which will then detect the edges of these regions, and then we will apply the genetic algorithm and by scanning further in the response space, try to find better quality class centers. This process is withdrawn until they are unable to find two sufficiently similar neighboring regions. The effectiveness of the proposed method was studied on a number of images. It is also compared by the K-means algorithm.

In the past decade, the fuzzy inference system (FIS) has seen explosive growth in popularity among the researchers and industrialists. The fuzzy classification system is based on the most knowing process of the FIS in order to map features as inputs to outputs classes. Generally, to make a fuzzy classifier decision it is necessary to select manually the best rules that require the intervention of experts. In the current paper, we propose an automatic method to select the best rules using associations models: Apriority and Filter Associations. The proposed method process in five steps: preprocessing (feature reduction), determining membership function for every input and output, nominal representation of the database basing on the membership functions, call for the association model to select the most important rules, and finally post-processing of the obtained rules. In this work, we lead to appropriate rules set tested on the iris data for the classification task. In addition, the proposed method is highly simple to be implemented to control the even complex system. Our system achieved promising results, which demonstrate the effectiveness of the proposed approach.

The automatic speech recognition (ASR) systems benefit from visual modality to improve its performance especially in noisy environments. By combining acoustic features with the visual features, audiovisual speech recognition (AVSR) system could be implemented. This paper presents a review on various existing and recent techniques for AVSR. A special emphasis was placed on recent AVSR system fusion technique, where the AVSR systems fusion stages (early, intermediate and late integration) are discussed with their corresponding models. The aim of this study is to discuss different AVSR approach and compare the existing AVSR techniques.

In this paper, an optimal sliding mode control based on neural network and genetic algorithm are designed for a DC motor velocity. The classical sliding mode control (SMC) can be used for the considered system. However, it presents some drawbacks of chattering, due to the higher needed switching gain in the case of large uncertainties. In order to reduce this gain, the neural network is used for the prediction of model unknown parts and hence enable a lower switching gain to be used. The neural network (NN) is used to improve the nominal model and then reduce the model uncertainties. This enables the sliding mode technique to be used without any chattering problems. The genetic algorithm is used in this study to optimize both, the learning rate of backpropagation algorithm used by the neural network and the variable switching gain of the SMC. The performance of the proposed approach is investigated in simulations by the comparison of the proposed approach with the classical sliding mode control technique.

This paper aims to build an interactive speaker-independent automatic Amazigh speech recognition system. The proposed system offers a methodology to extract data remotely from a distance database using the combined interactive voice response (IVR) and automatic speech recognition (ASR) technologies. We describe our experience to design an interactive speech system based on hidden Markov models (HMMs), Gaussian mixture models (GMMs) and Mel frequency spectral coefficients (MFCCs) based on ten first Amazigh digits and six Amazigh words. The best-obtained performance is 89.64% by using 3 HMMs and 16 GMMs.

In this paper, we introduce a new method of color image encryption. The principle of this approach is to encrypt any color image (called the original image), into a sequence of N images whose (N − 1) of them are generated randomly, and the Nth image is calculated from the original image and the (N − 1) generated images using a linear equation. This process will be repeated for each color image channel (red, green, and blue). Those (N − 1) generated images are used as a symmetric key in the decryption phase; therefore, they should be shared with the receiver privately or via a secure channel. Only the Nth image will be sent. To obtain the original image, we need all N images in the decryption phase. Our encryption system can resist brute-force attacks as well as statistical attacks. The results are justified by applying several safety criteria, such as correlation coefficient and histogram.

This paper presents a hybrid approach for 3D object reconstruction from multiple images taken from different viewpoints. The proposed method allows to obtain a complete and automatic reconstruction from limited number of images. It begins with a sparse 3D reconstruction based on camera self-calibration and interest point matching between images. The integration of sparse approach allows us to automatically estimate the projection matrices without using a turn-table (controlled environment) often used in the Shape from Silhouette approach. In addition, it offers the possibility of an accurate estimation of the initial bounding box of the object. This bounding box is discretized into voxels afterward. Then, the reconstruction process consists in using the image Silhouettes and the photo-consistency test to finally have a volumetric textured model that can be transformed into surface model by applying the marching cube algorithm. The experiments on real data are performed to validate the proposed approach; the results indicate that our method can achieve a very satisfactory reconstruction quality.

This paper presents a system that recognizes the lip movement for lip-reading system. Four lip gestures are recognized: rounded open, wide open, small open and closed. These gestures are used to describe visually the speech. Firstly, we detect the mouth region from frame using Viola–Jones algorithm. Then, we use DCT to extract the mouth features. The recognition is performed by a HMM which achieves a high performance of 84.99%.

Our article discusses existing methods in marker-based and markerless augmented reality and their evolution over time. Markers are the most optimal solution to solve the problem of calculating the pose for augmented reality and do not require powerful devices. Its simplicity, robustness and efficiency have a great advantage; the objects to detect are provided to the application and require to be always visible by the sensor. On the other hand, markerless augmented reality detects objects or characteristic points of a scene without any prior knowledge of the environment. This mechanism is more difficult to implement because it implements algorithms that are expensive in terms of computation time. The appearance and improvement of new devices have made it possible to exploit this technology and to detach from the presence of a marker in a scene and emerge to markerless methods.

Automatic speech recognition (ASR) in Amazigh speech, particularly Moroccan Tarifit accented speech, is a less researched area. Some efforts to develop ASR on Moroccan accented Amazigh speech in the clean environment have been studied in our previous works. In this paper, we analyze the effect of noise car at different signal-to-noise ratio (SNR) on Amazigh accented on the first ten digits with different decibel values. Various techniques are used for isolated speech recognition like hidden Markov model (HMMs) and Mel-frequency cepstral coefficients (MFCC). The experimental result presents that recognition rate of 88.22% in a clean environment, and 59.26% and 33.83% rates in noisy condition at SNR 10 dB and 20 dB, respectively.

We propose an algorithm for elimination of false correspondences from given putative point correspondences in couples of stereoscopic images based on corresponding circles (ICC) technique. In this paper, we propose a new method that reduces the cost of computing false correspondences. The aim of this method is mainly to take better advantage of the corresponding circles. The quantitative results demonstrate the efficiency and performance of our algorithm.

Distributed simulation with a virtual environment of quality is a qualified teaching solution, especially in the military field. A soldier's performance will be better if deployed preferably in a known combat space. This article deals, with a comparative study, the performance of two teams in training with two different simulation methods. One based on simulation in an imaginary virtual battlefield and the other on that existing in reality. The research and the results of this study represent the new approach of dynamic invention of real ground established, during the research carried out during the realization of a distributed simulator of combat for soldiers training on shooting techniques. This new method aims to provide trainees with a virtual combat space existing in reality and presenting the best results and performance and based on geographic information systems (GIS), some 3D modeling tools and 3D engines in this case the 3D Unity engine.

Image segmentation is an important step in any image analysis process. In the literature, there are two dual approaches. The contour segmentation approach consists in locating the object boundaries and the segmentation approach by region consists in partitioning the image into a set of regions. The best results of segmentation are obtained by cooperating these two approaches. They are more effective because the disadvantages of one method can be overcome by the advantages of another method. In an image processing system, the most important operation is image segmentation. To date, there is no universal method of image segmentation. Any technique is only effective for a given type of image, for a given type of application, and in a given computer context. Because of these constraints, the various image processing strategies that have been proposed have asserted their inadequacies and limitations. It is therefore perfectly normal to explore new horizons and find new methods that are more flexible and more effective. In this paper, we will propose and discuss a cooperative approach between a k-means unsupervised classification method and Canny's contour detection method to improve color image segmentation results. First, we will apply the contour segmentation process by Canny's algorithm, and then we integrate the results obtained by another unsupervised classification segmentation process, which is k-means. Experimental results show the strengths of our approach in terms of accuracy, simplicity and efficiency.

This paper presents an approach to reconstructing 3D objects based on the generation of dense depth map. From a two 2D images (a pair of images) of the same 3D object, taken from different points of view, a new grayscale image is estimated. It is an intermediate image between a purely 2D image and a 3D image where each pixel of this image represents a z-height according to its gray level value. Our objective therefore is to play on the precision of this map in order to prove the interest and effectiveness of this map on the quality of the reconstruction.

The main problem in the handwritten character recognition systems (HCR) is to describe each character by a set of features that can distinguish it from the other characters. Thus, in this paper, we propose a robust set of features extracted from isolated Amazigh characters based on decomposing the character image into zones and calculate the density and the total length of the histogram projection in each zone. In the experimental evaluation, we test the proposed set of features, to show its performance, with different classification algorithms on a large database of handwritten Amazigh characters. The obtained results give recognition rates that reach 99.03% which we presume good and satisfactory compared to other approaches and show that our proposed set of features is useful to describe the Amazigh characters.

Sentiment analysis is a very substantial area of research in our environment. Many studies have focused on the topic in recent years. It has rapidly gained interest due to the unusual volume of opinion-bearing data on the Internet (Big Social Data). In this paper, we focus on sentiment environment analysis from Amazon customer reviews shared by a machine learning based approach. This process starts with the collection of reviews and their annotation followed by a text pre-processing phase in order to extract words that are reduced to their root. These words will be used for the construction of input variables using several combinations of extraction and weighting schemes. Classification is then performed by a supervised Machine Learning classifier. The results obtained from the experiments are very promising.

Nowadays, companies and researchers are facing a real challenge to process amounts of data generated daily from different sources. To remedy this major problem, a range of technical solutions BIG DATA [1] is offered today. This article provides a detailed study of the main Big Data methods namely supervised and unsupervised methods as well as an interpretation.

In this work, we propose a method for reconstructing three-dimensional scenes from a sequence of images of which only one image is calibrated. The system is initialized by a calibrated image. The projection matrices of the other images in the sequence are estimated, from the first projection matrix (obtained by the calibration of the camera) and the matching between the images, by solving a system of linear equations. The intrinsic and extrinsic parameters of the camera used are estimated from the Cholesky decomposition of the projection matrix. Finally, obtaining the 3D scene is based on the Euclidean reconstruction of the detected and matched control points in the pairs of images.

In this work, we propose a method of cryptography of grayscale images. The principle of this approach is to encrypt any image (called the original image), into a sequence of N images whose (N-1) are generated randomly and the Nth image is determined from the original image and the (N-1) generated images. A key will be used in order to increase the security of the image transmitted to the receiver. The pixels of the original image are first converted into bits and then we use the shift bits operators to obtain a matrix, which will be used in the encryption procedure. All N images as well as the key used are needed in the decryption process, to obtain the source image. Experiments show the importance and quality of the approach that we proposed in terms of accuracy.

Per-pixel revolution mapping is an image-based modeling and rendering technique or IBMR (Image-Based Modeling and Rendering). It consists of generating very convincing revolved surfaces without polygonization effect, which are displayed interactively by the graphics cards. When it is about the interactivity or about the visual quality, this technique produces satisfactory results. When the revolved surface is created, we observed that the texture did not map in a correct way. In this paper, we presented an improvement of the revolution mapping technique to solve the problem related to the texture mapping.

The best texture mapping techniques that focus on the microrelief effect use iterative algorithms. The number of iterations must be determined in advance, but for real-time rendering, the number of iterations must be chosen dynamically to take into account the change of the depth scale and the viewing angle (viewing direction). In this paper, we presented a real-time rendering technique that allows a combined synchronization between depth scaling, viewing angle and number of iterations.

Scalability is an important and crucial issue which in routing protocols for Wireless Sensor Networks (WSNs). In this paper, we present an approach to achieving a balanced energy consumption rate using dynamic clustering to provide scalability in WSN. The proposed work in this paper is based on the dynamic clustering using k-means compared to LEACH (Low-Energy Adaptive Clustering Hierarchy), which is one of the most simple and effective clustering solutions widely deployed for WSN. The simulation results show that our proposed algorithm significantly improves high network scalability compared to LEACH.

Securing Data in the cloud based on Fully Homomorphic Encryption (FHE) is a new and potential form of security that allows computing on encrypted data without decrypted it first. However, a practical FHE solution is not available for implementation today. In this work, we propose a platform based on open source solutions to perform data computations (addition and multiplication) on encrypted form. In addition, taking account of efficiently and the security component, the most popular partially homomorphic encryption algorithms (RSA, Paillier and ElGamal) are studied to analyze the process times of encryption, decryption and computation of each algorithm. Furthermore, to compromise between performance and security, we need to study different key sizes and different data sizes as parameters.

The purpose of the work presented in this paper is to describe new method of 3D reconstruction from one or more uncalibrated images. This method is based on two important concepts: geometric constraints and genetic algorithms (GAs). At first, we are going to discuss the combination between bundle adjustment and GAs that we have proposed in order to improve 3D reconstruction efficiency and success. We used GAs in order to improve fitness quality of initial values that are used in the optimization problem. It will increase surely convergence rate. Extracted geometric constraints are used first to obtain an estimated value of focal length that helps us in the initialization step. Matching homologous points and constraints is used to estimate the 3D model. In fact, our new method gives us a lot of advantages: reducing the estimated parameter number in optimization step, decreasing used image number, winning time and stabilizing good quality of 3D results. At the end, without any prior information about our 3D scene, we obtain an accurate calibration of the cameras, and a realistic 3D model that strictly respects the geometric constraints defined before in an easy way. Various data and examples will be used to highlight the efficiency and competitiveness of our present approach.

In this work, we present an end-to-end system devoted to automatic recognition of printed Amazigh script in complex document images containing different languages such as Web images and natural scene images. To this end, text extraction from images is performed; the extracted text serves as input for a trained convolutional neural network (CNN) to identify its language. Finally, we proceed to the recognition of the Amazigh text script using a developed optical character recognition (OCR) system. The CNN reaches 99,12% of accuracy while the OCR system gets 99,93%. The obtained results seem to be very satisfactory.

We present our 3D models reconstructed based on geometric constraints. Used constraints in 3D reconstruction are frequently inaccurate. In fact, we detect an important set of theme, which one can we choose as input of reconstruction method? So, redundancies, cycles and inaccurate 3D model present a challenge to overcome. It is therefore necessary to “beautify” constraint and to elaborate an accurate 3D model before use in reconstruction systems. We present in our work a new method of 3D model elaboration. So, we remove redundancies; cycles and we bring a safe and final 3D model with an optimal freedom degree of estimated parameters. Our reconstruction algorithm first prioritizes the constraint detection. User can add them manually or through an automatic process. But the important thing is to avoid 3D modeling step which is very challenging. To detect and eliminate redundant and inconsistent constraints use freedom degree analysis. Results from our implementation show that the method can beautify and optimize recovered 3D models correctly at acceptable speed

The cone tracing is one of the best techniques for real time rendering of microreliefs. To find the intersection between the viewing ray and the microrelief, this technique use the empty space encoded in the form of top-opened cones. Cones are calculated in a preprocessing stage and are stored in a texture. There are two versions of this technique, the first one uses a conservative cone and the second uses a relaxed one. In this paper, we present the third version that uses the hybrid cone situated between the two already mentioned cones.

It is well known that speech recognition is a multimodal process which uses information not only from audio but also from vision. This paper describes our experience to design an audio visual speech recognition system, which relates the acoustic and the visual information in order to improve noise robustness of automatic speech recognition. The accuracy rate for face and mouth detection using Viola-Jones approach was satisfactory (reaches to 99% and 96.6% for face and mouth detection respectively).

Real-time rendering (Simulation, virtual Visits, 3D interactive, Games 3D) constitutes a major element of computer graphics, but the constraint of rendering speed does not allow to have high definition of all objects constituting the 3D scene. The traditional approach consists in using high definition meshes, having a very large number of vertices and triangles, but it is inappropriate for real-time rendering. Per-pixel extrusion mapping gives a serious alternative in this limitation by rendering the extruded surface without a polygonized silhouette. This paper takes advantage of programmable GPU's and related research as well as our approach the correction of the intersection points between the viewing ray and the surface generated by the extrusion of the binary shape and an extension of extrusion algorithm in order to create the outline extruded surfaces.

In last years, the facial biometry is coming back into education field; it is proposed to control student activities. In this paper, we propose a new face recognition system based on our Local Gradient Probabilistic Pattern (LGPP) and 2D-DWT. Firstly, the almost homogeneous and the picks areas are separate according to LGPP confidence interval. Secondly, the obtained images are decomposed using 2D-DWT in “LL, LH, HL and HH” sub-bands. Thereafter, we extract the features vector using 2D-PCA method applied on the approximation (LL-band). In classification phase, we compare between MLP, Bayesian Networks, SVM and KNN classifiers at features vector. The experimental results show that proposed system improves the recognition rate. Indeed, we reach a rate of 97 % for ORL and 98.8 % for Yale.

The Mean Shift procedure is a popular object tracking algorithm since it is fast, easy to implement and performs well in a range of conditions. However, classic Mean Shift tracking algorithm fixes the size and orientation of the tracking window, which limits the performance when the target's orientation and scale change. In this paper, we present a new human tracking algorithm based on Mean Shift technique in order to estimate the position, scale and orientation changes of the target. This work combines moment features of the weight image with background information to design a robust tracking algorithm entitled Scale and Orientation-based Background Weighted Histogram (SOBWH). The experimental results show that the proposed approach SOBWH presents a good compromise between tracking precision and calculation time, also they validate its robustness, especially to large background variation, scale and orientation changes and similar background scenes.

This paper presents a robust and important method for the self-calibration of a numerical camera using a tridimensional unknown scene. The particularity of our technique resides in the use of an arc of the circle and the localization of some of the matching points detected in the use of an ensemble of images in order to define in an automated way the intrinsic parameters, which are unknown and variable. Our approach is based on minimizing the constraints of the self-calibration of the camera with the projection of the two extremities of chord of each pair of images. It is obvious that in order to produce good results from the simulations we have made the choice of the right programming language would be of prime importance. Our main criteria were simplicity, stability and performing algorithms. The results have proven that the selection of the aforementioned criteria was successful.

There are a lot of difficulties facing a good handwritten Arabic recognition system such as the similarities of different character shapes and the unlimited variants in human handwriting. This paper presents a handwriting Arabic word recognition system. The objective of this approach is to propose an analytical offline recognition method of handwritten Arabic for rapid implementation. The first part in the writing recognition system is the preprocessing phase that prepares the data which serves to introduce and extract a set of simple statistical features by a window sliding along that text line from the right to left, then it injects the resulting feature vectors to the Hidden Markov Model Toolkit (HTK). In the recognition phase, the concatenation of characters to form words is modelled by simple lexical models, each word is modelled by a stochastic finite-state automaton (SFSA). The proposed system is applied to an "Arabic-Numbers" data corpus, which contains 47 words and 1905 sentences. These sentences are written by five different peoples.

In this work a robust technique of camera self-calibration with variable intrinsic parameters from unknown planar scene is presented. The main idea of our present method consists to use only two images of the scene by basing on the corresponding points between the two images which are numerous and easy to determine, the handling of only two images reduces remarkably the computation time of our method. From the projection of only two scene points and by using the image of absolute conic we will show that it's possible to obtain a set of self-calibration equations according to the intrinsic camera parameters. These equations are non-linear which required formulating of a non-linear cost function according to the intrinsic camera parameters which its minimization permits us to estimate these parameters in each image. The obtained experimental results on the synthetic and real images show the robustness and the reliability of our method.

In this paper, we address a problem of reconstruction of three-dimensional scenes from images taken by cameras, with varying parameters, from different views. This method is based on the projection of 3D points in the image planes. The relationships between the matches and the camera parameters are used to formulate a nonlinear equation system. This system is transformed into an objective function, which is minimized by a genetic algorithm to estimate the intrinsic and extrinsic camera parameters. Finally, the coordinates of 3D points of the scene are obtained by solving a linear equation system. The experiments on synthetic and real data show the quality of this work and the good obtained results.

Many applications in the field of computer graphics are becoming more complex and require more accurate simplification of the surface meshes. This need is due to reasons of rendering speed, the capacity the backup and the transmission speed 3D models over networks. We presented four basic methods for simplifying meshes that are proposed in recent years. The result obtained by the implementation of these methods will be the subject of a comparative study. This study aims to evaluate these methods in terms of preserving the topology and speed.

In this work, we propose a new local pattern for face recognition, named Local Gradient Probabilistic Pattern (LGPP). It is an extension of Local Gradient Pattern (LGP) that uses a very important result of probability theory; it is the law large numbers. In this direction, the distribution of the gray levels on a face image follows a law of probability, which is the sum of several normal laws. The current pixel will be evaluated by the confidence interval concept. The suggested model is merged with the most known algorithms of data analysis in the face recognition field, in particular the PCA, the LDA, the 2DPCA and the 2DLDA. The tests carried out on the ORL and YALE databases show the effectiveness of LGPP+2DPCA and LGPP+2DLDA systems. The experimental exactitude is of 96 %.

The Mean Shift algorithm for tracking the location of an object has recently gained considerable interest because of its speediness and efficiency. However, the appearance description using only color features cannot provide enough information when the target and its background have similar colors. In response to this problem, an improved human tracking system based on Mean Shift algorithm is incorporated in this paper. The proposed method combines color-texture features and background information to find the most distinguished features between the target and background for target representation. The experimental results show that the proposed method presents a good compromise between computational cost and accuracy; its performance is compared with recent state-of-the-art algorithm on Benchmark dataset and it achieved excellent results.

Structure from motion approach allows to recover both the 3D structure of the scene and the camera motion. It uses a global bundle adjustment by minimizing a non-linear criterion (often through the use of the Levenberg-Marquardt algorithm) to adjust the various entities initially estimated, which requires a long calculation time and can converge to a local solution due to a bad initialization. In this paper, we used the Structure from Motion approach for 3D object reconstruction from images taken by a single camera moving around the object. The proposed approach is based on the combination of the local bundle adjustment (LBA) and the global bundle adjustment (GBA) which is very useful to ensure effective and rapid convergence to the optimal solution. Our reconstruction system is initialized from two calibrated images. After each insertion of a new uncalibrated Image, we integrate a LBA to adjust the new estimated parameters and avoid as much as possible the accumulation of errors which can affect the system's stability. After the insertion of the last image, a GBA is performed to adjust as best as possible all the estimated entities already refined locally (3D points and camera parameters). Experimental results show the reliability and rapidity of the proposed approach compared to the classical approach.