Earticle

Computer-aided diagnosis (CADx) is used to help radiologists in interpretation mammograms and is usually used as a second opinion by the radiologists. Improving CADx increases the treatment options and a cure is more likely. The main objective of this research is to enhance and introduce a new method for feature extraction and selection in order to build a CADx model to discriminate between cancers, benign, and healthy parenchyma. For feature extraction, we use both human features, which are obtained by Digital Database for Screening Mammography (DDSM), and computational features. For computational feature extraction, we enhance and use two pre-existed feature extraction methods, which are the Run Difference Method (RDM) and the Spatial Gray Level Dependence Method (SGLDM). Then, we evaluate and introduce a new method for feature selection by running both of forward sequential and genetic algorithm search methods individually. Later we evaluate the results. Experimental results are obtained from a data set of 410 images taken from DDSM for different types. Our method select 14 features from 65 extracted features. We used both Receiver Operating Characteristics (ROC) and confusing matrix to measure the performance. In training stage, our proposed method achieved an overall classification accuracy of 94.6%, with 95.2% sensitivity and 84.8% specificity. In testing stage, our proposed method achieved an overall classification accuracy of 87%, with 88.6% sensitivity and 78.6% specificity.

Image zooming have become an important topic in image processing and analysis. New image zooming is introduced in this paper based on K-Space transformation technique. The main aim of the proposed scheme is to achieve high zooming factor without much effect on the image quality. Simulation results on real digital images are given to show effectiveness and reliability of the proposed algorithm.

Content-based medical image retrieval is an important tool for doctors in their daily activity. In this paper, we propose a novel image retrieval framework to combine visual concept and local features. To obtain visual semantic representation of the image, we first construct a graph model by feature distance and density similarity, and then a graph-based semi-supervised learning method is applied to get the membership degree of query images. Meanwhile, the dense SIFT feature of the image patches is extracted and described by bag of visual words as local features. Besides, we design a similarity measurement based on visual concept and local feature rather than using low level features only. We evaluate the proposed algorithm in ImageCLEFmed dataset. The results demonstrate that our method represents the visual semantic of images effectively, and compares favorably to state-of-the-art approaches based on single low level features in retrieval performance.

This paper investigates the possibility of exploiting facial skin texture as a source of biometric information to facilitate automatic recognition of individuals. Such ability may be particularly important in circumstances when a full view of the face may not be available. The proposed algorithm automatically segments the forehead region and divides it into non-overlapping patches. Two state-of-the-art families of texture feature extraction approaches, namely Gabor wavelet filter and Local Binary Pattern operator, are compared for extracting features from these patches which are classified using a k-NN classifier. The identification and verification performance is evaluated for different patch sizes using the XM2VTS database. For the verification experiments an EER of 0.065 using Gabor features and 0.083 using LBP features is obtained for forehead regions with pure skin. Additionally a novel classifier is presented for automatically detecting pure skin patches in the forehead region.

In this paper a novel quick automatic method is proposed for electrocardiogram (ECG). Signal classification to three classes include: the normal heart beats from the left bundle branch block (LBBB), right bundle branch block (RBBB), and paced beats. After noise reduction using wavelet threshold, appropriate features are extracted from the time-voltage waves including P, Q, S, and T waves in ECG signals. Novelty of this work is utilization of fast decision based on non-parametric statistical classifier and Multi Features Data Fusion (MFDF) strategy. Two stages of MFDF include feature classification into normal and abnormal categories. Based on decision template, first stage, and second part are voting and weighting the procedure. Post processing block is added for impulsive noise reduction in order to improve the results. We emphasized on the performance and efficiency of the optimized presented algorithm and minimum cost of system learning. The accuracy of final results is reliable and well performed.

A new method, 2DPCA-GMM of tracking and segmentation in the dynamic environment of objects is proposed in this paper. The method attempts to link the Gaussian mixture model, (GMM) with the method of two dimensional principal component analysis (2DPCA) and apply Kalman Filtering (KF) for tracking. In this context, the aim of the paper is to tackle tracking of moving object based on 2DPCA-GMM together with Kalman prediction of the position and size of object along the image’s sequence. The obtained results successfully illustrate the tracking of a single moving object as well as multiple moving objects based on segmentation generated by 2DPCA-GMM.

In this paper, a fast and memory efficient version of set partitioned embedded block (SPECK) image coder is proposed. Due to use of linked lists, SPECK algorithm, requires large run-time memory, making it unsuitable for memory constrained portable multimedia devices and WMSN’s. The proposed coder replace linked list with fixed size static memory, to keep track of set partitions and coding information and use few markers to facilitate coding. Elimination of linked lists also reduces the memory access time and time involved in addition /deletion in memory, thereby making it faster than the original SPECK. Simulation results show that the proposed algorithm outperforms SPECK image coder in terms of memory requirement while retaining the coding efficiency and scalability property of the original SPECK. Proposed coder requires 0.75 bit per pixel state memory which is 9.38 % of the memory required to store the image. Therefore proposed NLSK coder is suitable for resource constrained portable hand held devices and WMSN’s.

In a learning-based super-resolution algorithm, suitable prior from the training database is a key issue. A novel face hallucination algorithm based on shape clustering and subspace learning for adaptive prior is proposed in this paper. We define face shape metrics with point distribution model by Hausdorff Distance, then a framework of adaptive prior and subspace learning is proposed to enhance the performance of surveillance face super-resolution. Linear regression is used to learn the relationship between low and high image systhesis coefficients. Experiments show that the face super-resolution algorithm based on shape classification can improve the subjective and objective quality of the input low-resolution face images and outperform many state-of-the-art global-based face super-resolution methods.

Many important applications have used novelty detection in order to reduce redundant and non-relevant information presented to users of the document retrieval systems. In this study, sentence level information patterns are proposed for enhancing the novelty detection for Arabic text documents. Two models based on sentence level information patterns are suggested and compared; the first one is based on sentence length while the second one is based on opinion patterns. Experimental results have showed that both of the proposed models; Length Adjusted (LA) model and Length and Opinion Adjusted (LOA) model, can significantly improve the performance of novelty detection for Arabic texts, in terms of precision at top ranks. Better results were provided by LA model over LOA model. This shows that the sentence length is more important for enhancing the novelty detection than other suggested sentence level information patterns (e.g. opinion patterns).

The variation of facial appearance due to the viewpoint or pose obviously degrades the accuracy of any face recognition systems. One solution is generating the virtual frontal view from any given non-frontal view to obtain a virtual gallery/probe face. As the state-of-the-art face recognition algorithm, linear regression computes a reconstruction matrix from the images of each subject and then approximates the probe face image by using the reconstruction matrix, but the performance of this linear algorithm is limited due to the nonlinear structure of the face images which is caused by variations in illumination, expression, pose and occlusion. Following this idea, in this paper, we propose an efficient and novel locally kernel-based nonlinear regression (LKNR) method, which generates the virtual frontal view from a given non-frontal face image. Because of the high (even infinite) dimensionality of the nonlinear transformation functions, it is infeasible to directly calculate the corresponding reconstruction matrix and therefore is unable to approximate explicitly the probe image. So, with the help of kernel functions, we overcome to this mentioned problem by embedding the nonlinear regression in the stage of computing the reconstruction matrix from the non-frontal input face and non-frontal face database. The comparison of the proposed method with locally linear regression (LLR) and eigen light-field (ELF) methods is also provided in terms of the face recognition accuracy. Experimental results show that the proposed method outperforms two other methods in terms of robustness and visual effects.

This paper is a humble step towards investigating feasibility of implementation of Code Excited Linear Prediction algorithm on Adaptive Multi Rate Wideband coder over audio signals. Proposed coder offers inherent capability of adaptively change its bitrates with variation in received C/I ratio depending upon channel conditions. Though proposed WB coder exhibits nine different bitrate modes (between 6.6kbps and 23.85kbps), its practical implementation reveals that only two upper most bitrates (which are 23.05kbps and 23.85kbps) have been found suitable for audio transmission and recovery over WB link. An e-test bench using MATLAB is created to implement proposed WB coder and series of simulations are carried out to judge the performance of implemented coder for audio signal using Subjective (Mean Opinion Scores) and Objective (Perceptual Evaluation of Audio Quality- Objective Difference Grade) analysis. Simulation results clearly advocate that it is possible to reproduce audio signals with comparable quality when implemented with upper most two bitrates of AMR WB coder. It is also evident from the obtained simulation results that PEAQ and MOS for chosen audio signal are also quite comparable and satisfactory.

There are many traditional pitch detection methods, but most of them can’t perform perfectly for different speakers, applications and environmental conditions. For this reason, a pitch detection method based on multi-feature is proposed. Firstly, the speech signals are pre-filtered. Secondly, the speech signal pre-filtered is segmented into syllables. Finally, the pitch period is obtained by wavelet transform and the maxima selected. Experiments show that this method can increase the performance of pitch detection in both clean and noisy environment compared with weighted wavelet method.

As we know the population of city and number of cars is increasing day by day. With increasing urban population and hence the number of cars, need of controlling streets, highways and roads is vital. In this paper, a system that estimates the size of traffic in highways by using image processing has been proposed and as a result a message is shown to inform the number of cars in highway. This project has been implemented by using the Matlab software and it aims to prevent heavy traffic in highways. Moreover, for implementing this project following steps must be considered: 1) image acquisition 2) RGB to grayscale transformation 3) image enhancement and 4) morphological operations. At first, film of highway is captured by a camera has been installed in highway. Then, the film comes in the form of consecutive frames and each frame is compared with the first frame. After that, the number of cars in highways is specified. At the end, if the number of cars is more than a threshold, a message is shown to inform the traffic status. By this message we can predict the need to reduce the size of traffic carried. Experiments show that the algorithm will work properly.

The use of characteristics of iris and pupil are very useful in wide range of applications. There have been many studies about detecting pupil and iris, however most of these studies focused on images which are taken under controlled illumination conditions. In this paper, we focus on images in which are taken under uncontrolled illumination using compact digital cameras. We propose an efficient method for detecting pupil and iris accurately. Our method is based on Hough Circle Transform by fixed radius. This algorithm combines face recognition, edge detection and characteristics of LAB images. We implement our method into 125 face images which are taken by compact digital camera under uncontrolled illumination, varying pose and expression conditions of human face. The result shows that our algorithm has performance about 82%.

Hyper-spectral Atmospheric Infrared Sounder AIRS has many channels. Taking into account the correlation between channels, different detection purposes and the timeliness of assimilation, channel selection is necessary. Principal component-dual-zone-stepwise regression method is proposed in this paper and points the day and night for channel selection. The contribution rate of channel observation brightness temperature to the objective function is controlled by channel observation error, variational data assimilation system is generally given observation error and then in the process of minimization objective function remains unchanged, the paper adopt the observation error re-estimated and then provide observation error information to the variational data assimilation. Study effects of channel selection and the observational error allocation on the assimilation of AIRS data in this paper.