Pitch frequency estimation is one of the important issues in speech processing. Detection of formant frequencies is also a basic step in speech processing. Formant frequencies are essentially, resonance frequencies of vocal tract. These frequencies vary among different persons and words but they are within certain frequencies. Practically, calculation of three formants is sufficient for coding and other processes. The most prevalent method for estimating formants is using cepstrum. In this study a new method has been presented based on wavelet transform and filter bank notion.

In this paper, a new robust orientation estimation for Coherence Enhancement Diffusion (CED) is proposed. In CED, proper scale selection is very important as the gradient vector at that scale reflects the orientation of local ridge. For this purpose a new scheme is proposed in which pre calculated orientation, by using robust orientation estimation, is used to find the correct true local scale. From the experiments it is found the proposed scheme is working much better in noisy environment as compared to the traditional Coherence Enhancement Diffusion

Glaucoma is considered to be one of the major causes of blindness but it was hard to diagnose it in early stages. Segmentation of Retinal blood vessels in the ophthalmological images is the most valuable parameter to diagnose glaucoma. This paper uses image processing techniques to segment the vasculature inside the optic disc to calculate its ratio in ISNT quadrants. The ratio of blood vessels in ISNT quadrants is evaluated to check whether blood vessels are being nasalized, obeys or violates the ISNT rule. The proposed method is tested on 50 images taken from DMED, MESSIDOR and FAU to examine the nasalization of vessels in retinal images.

H.264/AVC has been widely applied to various applications. However, a new video compression standard, High Efficient Video Coding (HEVC), had been finalized recently. In this work, a fast transcoder from H.264/AVC to HEVC is proposed. The proposed algorithm includes the fast prediction unit (PU) decision and the fast motion estimation. With the strong relation between H.264/AVC and HEVC, the motion vectors (MVs), residuals, and modes from each coding block of H.264/AVC can be reused to predict the current encoding PU of HEVC. Furthermore, the MV variance from H.264/AVC is calculated to decide the search range of PU and also to reduce the prediction mode. Simulation results show that the proposed method can save up to 53% of the encoding time and maintains the rate-distortion (R-D) performance for HEVC.

Today several methods for automatic identification of chromosome images are represented in order to improve the AKS (Automatic Karyotype System) diagnosis capabilities. Healthcare researchers have spent a lot of effort to deal with main challenges in this area that are accuracy rate and time consuming to make classification results feasible for clinical applications. Firstly this paper presents an efficient preprocessing step using FFT on gray-scale images of 22 pairs of human autosome chromosomes to minimize noise and obtain well defined features with polarity free. Secondly a powerful support vector machine with multiple kernels have been employed that to identify single chromosomes. Experimental results demonstrate that the best accuracy is 97.8% on Copenhagen data set, and the worst one reported is about 95%, which is exceeded more existing methods.

Edge detection is a very important aspect of image processing. Edge detection refers to the algorithms which aim at identifying point in a digital image at which the image brightness changes significantly. The need of edge detection is to find the discontinuities in depth, surface orientation, changes in material properties and variations in scene illumination. This paper proposes a novel edge detection technique which uses the logical operations to efficiently detect the edges with lesser computational complexity and time. Experimental results show that the proposed algorithm inhibits noise effectively, improves the detection accuracy, and possesses encouraging robustness when applied to different images, in comparison with several classical edge detection algorithms.

Wireless Channels intended to achieve very high levels of spectral efficiency generally utilize very opaque QAM constellations. A Power Adaptation Algorithm with minimum power(MPAA) is proposed using 4-QAM to transmit over AWGN channel. This algorithm optimizes RootMean Square Error with better Peak Signal to Noise Ratio and good bit error rate using 4- QAM modulation compared over Conventional Power Adaptation Algorithm (CPAA).

Heartbeat and respiration dynamics, characterized by the series of intervals between two successive beats and breath, respectively, are associated with each other. In this paper we performed experiments to validate the existence of a nonlinear relationship between the two signals. Heartbeat and respiration signals were obtained from a database of physiological signals, and from there Heart rate variability (HRV) and Interbreath Interval (IBI) signals were derived and used in training feed-forward neural networks in an attempt to capture the relationship between the two dynamics for each subject. Analysis includes comparing the RMS and crosscorrelation coefficients of the original respiration signal with the predicted time series, and performing the same operation for random signals with similar fractal characteristics as original RR signals. The higher cross-correlation coefficients of the predicted signal to the target signal for the young patients supports our stand that there is a non-linear relationship between the two physiological dynamics.

We present a 3D real time image matching system based on the skeleton tracking module of Kinect Sensor. To determine an optimal correspondence between real time and standard image data, we use a dimensional angular representation as a descriptor of the skeleton designed for recognition robustness under noisy input. And we compute a distance metric to evaluate the difference in motion between the poser's body and the template. The grading formula for image matching is adaptively based on our experimental data, which makes our system stands a high accuracy under input noise from the real time depth sensor.

In this paper, we propose a novel global threshold-based active contour model, it employs a new edge-stopping function that controls the direction of the evolution and stops the evolving contour at weak or blurred edges. The model is implemented using selective binary and Gaussian filtering regularized level set (SBGFRLS) method. The method has a selective local or global segmentation property. It selectively penalizes the level set function to be a binary function. This is followed by using a Gaussian function to regularize it. The Gaussian filters smooth the level set function and afford the evolution more stability. The contour could be initialized anywhere inside the image to extract object boundaries. The proposed method performs well when the intensities inside and outside the object are homogenous. Our method is tested on synthetic, medical and Arabic-characters images with satisfactory results.

Image segmentation is the most challenging issue in computer vision applications. And most difficulties for crops management in agriculture are the lack of appropriate methods for detecting the leaf damage for pests' treatment. In this paper we proposed an automatic method for leaf damage detection and severity estimation of coffee leaf by avoiding the defoliation. After enhancing the contrast of the original image using LUT based gamma correction, the image is processed to remove the background, and the output leaf is clustered using Fuzzy c-means segmentation in V channel of YUV color space to maximize all leaf damage detection, and finally, the severity of leaf is estimated in terms of ratio for leaf pixel distribution between the normal and the detected leaf damage.

Lane boundary detection on sub-urban streets based on images acquired by a video capturing device is a challenging task. This is mainly due to the problem of estimating the complex geometric structure of the lane boundaries, poor quality of lane markings resulting due to wear, occlusions from traffic and shadows caused by road side trees and other constructions. Most of the existing techniques employ a single visual cue in lane boundary detection which is bound to work only under certain conditions where there are clear lane markings. It gives better results when no other on-road objects are present. In this paper, we introduce a novel lane boundary detection algorithm that specifically targets the aforementioned issues. The method addresses these issues by integrating two visual cues; the first visual cue is based on extraction of stripe like features found on lane lines by a two dimensional symmetric Gabor filter. The second visual cue is based on texture characteristic that employs the entropy measure of predefined neighborhood around a lane boundary line. These visual cues are then integrated by a rule based classifier which incorporates a modified sequential covering algorithm, thus improving the robustness. To separate lane boundary lines from other similar features found elsewhere, a road mask is generated depending on the road chromaticity values estimated from CIE L*a*b* color transformation. After this process, extraneous points around lane boundary lines are removed by an outlier removal procedure based on studentized residuals, followed by modeling the lane boundary lines with Bezier spline curves. The algorithm is implemented and an extensive experimental evaluation is carried out on sub-urban streets, confirming the approach introduced in this paper.

In this paper we propose an mean shift Kalman object tracking algorithm for video surveillance which is based on the mean shift algorithm and the Kalman filter. The classical mean shift algorithm for tracking in perfectly maintained conditions constitutes a good tracking method. This was based on color to predict the location of the object in the video frame. However, in real cluttered environment this fails especially under the presence of noise or occlusions. In order to deal with these problems this method employs a Kalman filter to the classical mean shift algorithm to enhance the chance of tracking accuracy especially when the object disappears from the scene, the algorithm can still track the object after it comes out. In this algorithm a search window was used in order to increase the performance of the algorithm. The experimental results verifies the ability of the mean shift Kalman object tracking algorithm which can locate the target object correctly even in difficult situations when the target is occluded.

In this paper, classifying and indexing hierarchical video genres using Support Vector Machines (SVMs) are based on only audio features. In fact, these segmentation parameters are extracted at block levels, which have a major benefit by capturing local temporal information. Our contribution in this study is to present a powerful combination between the two employed audio descriptors that leads to a higher classification accuracy. Our purpose is to classify a big YouTube dataset that includes three multi-Arabic dialects video genres and five sub-genres: several sports analysis and various matches categories (foot-ball, basket-ball, hand-ball and volley-ball), both studio and fields news scenes over and above various multi-singer and multi-instruments music clips. Our database was carried out on over 18 hours of video span yielding a classification accuracy of 98,5% for genres and 97% for sub-genres. Finally we discuss the performance of SVM kernels applied onto our video dataset.

Recognition of handwritten Arabic text awaits accurate recognition solutions. There are many difficulties facing a good handwritten Arabic recognition system such as unlimited variation in human handwriting, similarities of distinct character shapes, character overlaps, and interconnections of neighboring characters, their position in the word; Arabic letters are drawn in four forms: Isolated, Initial, Medial, and Final. The typical Optical Character Recognition (OCR) systems are based mainly on three stages, preprocessing, features extraction and recognition. Each stage has its own problems and effects on the system efficiency which is the time consuming, resources using and the recognition errors. There are many feature extraction methods for handwritten letters. In this paper, an efficient approach for the recognition of off-line Arabic handwritten characters is presented. The approach is based on structural Statistical and Morphological features from the main body of the character and also from the secondary components. Evaluation of the importance and accuracy of the selected features is made. An off-line recognition system based on the selected features was built. The system was trained and tested with dataset. We use the popular Feed Forward Neural Network for classification to find the recognition accuracy. The proposed algorithm obtained promising results in terms of accuracy (success rate of 100% for some letters at average 88%) as well as in terms of time consuming. In Comparable with other related works we find that our result is the highest among others.

The disadvantages of domestic and oversea shadow image edge detection algorithms are analyzed.A novel shadow detection algorithm based on edge growing and rough set theory and subsequent solution is proposed. We describe how to detect image edge using condition attribution of rough set in this paper. Also, the method of thinning and connection for shadow edge using edge growing from the edge nodes is proposed. As can be seen from the experimental analysis, the method we proposed has better performance in edge detection and image segmentation.

The Acoustic echo problem appeared very long time ago at the time of the phone invention, at that time, this problem was solved by means of analog technique. If we consider a case where there is both microphone and loudspeaker, for example in a conference the speech playing on the speaker captured by the microphones and again played in the speaker, this played voice (transmitted signal) is delayed version of the original one. This may causes reverb, frequency filtering and attenuation, is lowering down the voice quality to an unacceptable level. So to have a correctly received signal we have to use Acoustic Echo Cancellation (AEC), which can simulate the transmitted signal and therefore abstract the optimized signal.

This paper proposes a Chinese Abacus DAC for high speed and low power applications like audio and video applications. This circuit of DAC uses resister strings to get a good analog output. The designed DAC uses the algorithm of abacus. Instead of using binary code, here we use abacus code to control the switches. So the complexity and the area will be reduced automatically. The 8-bit DAC is comprised of 12 resistors and 24 NMOS switches. The 8-bit Abacus resistor DAC requires 12 resistors and 24 switches. The 8-bit resistor-string DAC requires 255 resistors and 256 switches. The most important advantages are that the numbers of both resistors and switches are all reduced effectively. The simulation environment uses 1 um process technology.

Humans' ability to detect and locate salient objects on images is remarkably fast and successful. Performing this process by using eye tracking equipment is expensive and cannot be easily applied, and computer modeling of this human behavior is still a problem to be solved. In our study, one of the largest public eye-tracking databases [1] which has fixation points of 15 observers on 1003 images is used. In addition to low, medium and high-level features which have been used in previous studies, SIFT features extracted from the images are used to improve the classification accuracy of the models. A second contribution of this paper is the comparison and statistical analysis of different machine learning methods that can be used to train our model. As a result, a best feature set and learning model to predict where humans look at images, is determined.

Modern multimedia tools heavily depend on image manipulation. In many cases, manipulation of images is specific to certain directions or angles. Generally, the images are manipulated using software routines, which are time consuming, compared to manipulation using hardware techniques. This paper proposes hardware architecture, based on two dimensional cellular automata, for manipulating images in certain specific directions and angle. The proposed architecture can be easily implemented using VLSI technology.

It can speed up the efficiency of modeling and save costs to extract the vector line of buildings automatically based on aerial images. The method used in this paper extracts DSM and DEM with aerial images, and then gets the ground object by subtracting DEM from DSM. Considered the geometric surface features, spectral, and texture ,buildings can be extracted by segmenting aerial images. Finally this paper conducts a three-dimensional reconstruction using building outline and accuracy assessment. The results show that this method can extract the building outlines rapidly and accurately.

This work proposes a novel system for Violent Scenes Detection, which is based on the combination of visual and audio features with machine learning at segment-level. Multiple Kernel Learning is applied so that multimodality of videos can be maximized. In particular, Mid-level Violence Clustering is proposed in order for mid-level concepts to be implicitly learned, without using manually tagged annotations. Finally a violence-score for each shot is calculated. The whole system was trained on a dataset from MediaEval 2013 Affect Task and evaluated by its official metric. The obtained results outperformed its best score.

Various texture, shape, boundary features have been used previously to classify regions of interest in radiological mammograms into normal and abnormal categories. Although, bag-of-phrases or n-gram model has been effective in text representation for classification or retrieval of text, these approaches have not been widely explored for medical image processing. Our purpose is to represent regions of interest using an n-gram model, then deploy the n-gram features into a back-propagation trained neural network for classifying regions of interest into normal and abnormal categories. Experiments on the benchmark miniMIAS database show that the n-gram features can be effectively used for classification of mammograms into normal and abnormal categories in this way. Very promising results were obtained on fatty background tissue with 83.33% classification accuracy.

Automatic speech recognition (ASR) has been extensively studied during the past few decades. Today, most of the ASR system based on statistical modelling, and HMM is the most popular one among them. Now a days, performance of ASR become one of the major bottleneck for its practical use. Literature of ASR shows that optimal performance could be achieved by using Gaussian mixture hidden markov model (HMM) but choice of Gaussian mixture is arbitrary with little justification. If we use the different number of Gaussian mixture then we get different results. In this paper, we compare the performance of continuous Hindi speech recognition by using different number of Gaussian mixture. The aim of this paper is to investigate the optimal number of Gaussian mixture that exhibits maximum accuracy in the context of Hindi speech recognition. HMM toolkit HTK 3.4.1 is used for the implementation of this system, in which Mel frequency cepstral coefficient (MFCC) is used as a feature extraction technique. The experimental results shows that the maximum performance of the proposed system is achieved when we use four component Gaussian mixture HMM model.

Speech to Text recognition is the ability of a machine to recognize the human speech and convert in to text sequence. In this paper, we compare the performance of isolated word, connected word, and continuous speech recognition system with different vocabulary sizes. Hidden Markov Model toolkit HTK 3.4.1 is used to develop the system. For feature extraction, Mel Frequency Cepstral Coefficient (MFCC) and Perceptual Linear Prediction (PLP) both are used in this paper. The aim of this paper is to build a high performance speech recognition system for Hindi language. Hidden Markov Model (HMM) and Gaussian Mixture Model (GMM) are used at the back-end of our proposed system. The system is trained for 100 Hindi words and each word 10 utterances have been recorded for training of the ASR system. The experimental result shows that the overall accuracy of proposed system with 100 word dictionary size is 95.40%, when we use the combination of MFCC and GMM for automatic speech recognition (ASR) system.

The internet community has been benefitting tremendously from the works of various researchers in the field of Natural Language Processing. Semantic orientation analysis, sentiment analysis, etc. has served the social networks as well as companies relying on user reviews well. Flame identification has made the internet less hostile for some users. Spam filtering has made the electronic mail a more efficient means of communication. But with the incessant growth of the internet, NLP using machine learning working on massive sets of raw and unprocessed data is an ever-growing challenge. Semi-supervised machine learning can overcome this problem by using a large set of unlabeled data in conjunction with a small set of labeled data. Also, focusing on developing NLP systems that can contribute to developing a unified architecture could pave the way towards General Intelligence in the future.

In this paper, we propose a framework for Age Estimation which uses a correlated ageing pattern to rank images and makes necessary inferences from the image ranks to estimate the exact age of images. We use AAM and LBP as complementary feature extraction techniques for extracting facial features in low dimensionality. Our correlated ageing pattern model learns the ageing patterns of different individuals across ages and uses these to determine an agerank for each image. Subsequently, the learned age rank of a reference image set is used to determine the ranks of test images in order to deduce relevant inferences for age estimation. Our approach is significantly different from the previous ranking approaches in that it determines age ranks that do not only represent the correlation of ages of different individuals but also the correlation of ageing patterns of different individuals. Our initial findings look promising with the intuitive manner with which we employ correlated ageing patterns.

Ageing is a natural phenomenon and it is part of our daily life process; the Facial Ageing (FA) process has been of a great interest to many researchers and some firms like airports and police departments; this is due to the fact that the face appearance changes as people age resulting in difficulties identifying certain individuals. In this paper, two-dimensional wrinkle maps will be used in the design of a three-dimensional system for the purpose of facial wrinkles simulation and prediction. Our findings will challenge many commercial softwares in the innovation of the techniques in setting solid grounds to generate real-time 3D wrinkles that can be used later for various reasons that may include security, 3D facial ageing simulation systems, facial animation, etc. The 2D binary wrinkles will be mapped on the corresponding 3D face models using the generated outlined images. NURBS curves will then be projected on those wrinkles to form a three-dimensional wrinkle map. The coloured wrinkle map, as well as some parameters, will be combined together in an algorithm to predict the appearance of the individual wrinkles in every age group that are divided into decades, starting from the age of 20. The novelty of the adopted procedure in comparison to the previous works, is the new elements that have been integrated and collaborated to boost accuracy and to generate a more realistic outcome.

Object recognition, the task of finding a given object in an image, is an important problem which may help in numerous applications such as face detection, pedestrian detection, vehicle tracking, global robot localization, gesture recognition, optical character recognition etc. To achieve better performance in object recognition, carefully extracted features in addition to a suitable model (i.e. combination of features) are needed. This paper gives an overview of the most used local and global features as well as combinations of them. Recently, researchers proved that combining both local and global features has a drastic impact on the performance of the recognition task. Also, the paper emphasizes that using the depth information improves the object recognition system accuracy. This has recently been exploited with the availability of cheap RGB-D sensors such as Microsoft Kinect.

Image reconstruction is an inverse problem which can be formulated using quadratic objective functionals (Least Square fittings or L2 norms) and absolute values summations (L1 norms). The L1 and L2 norms can be independently applied over the data mismatch and the regularization terms (image term) of an inverse problem. In this manuscript, we investigate weighted L1 and L2 norms in constituting a general inverse problem and reconstructing images using Primal-Dual Interior Point Method (PDIPM). We propose a generalized inverse problem to independently mix the smooth properties of the L2 norm based objective functionals with the blocky effect of the L1 norm based objective functionals on a element by element basis through a weighting strategy. In our implementation, we use Electrical Impedance Tomography (EIT) as an instance of an ill-posed, non-linear inverse problem. We investigate the effectiveness of different combinations of weighted L2 and L1 norms in dealing with measurement uncertainties, such as measurement noise and data outliers, using both EIT simulated data, and EIT human lung data. The simulated data is produced for a 2D circular phantom and EIT conductivity images are reconstructed. The first clinical results of applying weighted L1 and L2 norms to reconstruct image of EIT lung data using a 2D thorax-shape mesh are reported.

High contrast image reconstruction is highly desirable in many applications including medical and industrial imaging applications. One approach for such high contrast images is Level Set based reconstruction methods (LSRM). In this paper, we propose a novel variant of the LSRM and evaluate it against traditional approaches. The proposed LSRM allows for all possible combinations of the L1 norms (sum of absolute values) and the L2 norms (least square fittings) on the inverse problem terms. To show the implementation of the derived LSRM, we use an ill-posed, non-linear inverse problem known as Electrical Impedance Tomography (EIT). The image reconstruction results of the proposed LSRM are compared to those of the four state of the art regularization methods: Gauss-Newton (GN) with Tikhonov regularization term, GN with NOSER algorithm, Total Variation (TV), and the PDIPM. According to our results, the proposed LSRM produces more robust results in the presence of high level of noise (addition of 60dB Gaussian noise) and strong outliers (loss of measurement data) when compared with the competing methods. We also apply the proposed LSRM over the EIT human lung data and showthat the proposed LSRM produces physiologically plausible images.

Advances in technology have diode to development of varied sensing, computing and communication devices which will be plain-woven into the physical setting of our daily lives. Such systems alter on-body and mobile health-care watching, will integrate data from completely different sources, and might initiate actions or trigger alarms once required. Finally, we tend to gift our results, demonstrate the practicability of our projected techniques and description the longer term directions. With recent developments within the wireless networks field, new and innovative medical applications supported this technology square measure being developed within the analysis likewise as industrial sectors. This trend has simply started and that we predict wireless networks square measure reaching to become an integral a part of medical solutions because of its advantages in reducing health care prices and increasing accessibility for patients likewise as networks within the medical field and discuss the problems and challenges. We have conjointly tried to spot a number of the standards in use. Another contribution because of this paper is that the identification of innovative medical applications of wireless networks developed or presently being developed within the analysis and business sectors. Within the finish we tend to conjointly refer the longer term trends during this field.

Melodic similarity is very important for analysis and classification of classical music. The difficulties to measure the melodic similarity are mainly structural complexity and melodic variations. It is more difficult to use machine learning techniques to measure it automatically. In this paper we use a hybrid of two methods: numbered musical notation and dynamic time warping. Several classic music pieces are used to demonstrate effectiveness of our method. This method can be directly extended to measure the melodic similarity of the other types of music

Acoustic signals in a shallow ocean environment are severely distorted due to the time-varying and inhomogeneous nature of the propagation channel. In this paper, a state-space model is introduced to characterize the uncertainties of the shallow ocean and a Rao-Blackwellized particle filter (RBPF) is developed to estimate the model parameters. Since both modal functions and horizontal wave numbers of the channel are assumed unknown, the state-space model has a high nonlinearity and high dimensionality. As the modal functions are linear with the measurements conditioning on the horizontal wave numbers, a Kalman filtering (KF) is employed to marginalize out the modal functions. Hence only the horizontal wave numbers need to be estimated by using a PF. Simulation results show that the proposed RBPF algorithm significantly outperforms the existing approaches