Earticle

Vehicle plate recognition is an effective image processing technique used to identify vehicles' plate numbers. There are several applications for this technique which expand through many fields and interest groups. Vehicle plate recognition may be used as a marketing tool, for purposes of traffic and border control, for law enforcement, and travel. Many methods have been proposed to facilitate this technique. This study proposes an edge-detection method to enable a Plate Recognition System through practical situations, such as various environmental or meteorological conditions. Image processing tools are used to scan the plate area, resize it, and convert it toward a gray scale prior to filtering the image in order to remove small objects. The obtained objects are identified such that the numbers object is recognized. The details of the obtained image are controlled through the standard deviation of the Gaussian filter (sigma).

Geometric and skin face characteristics were used, in this paper, to localize face in images. In first stage, well known horizontal symmetric characteristic of face is exploited to localize the vertical symmetry axis of the face, then to delimit its vertical position in the image. In second stage, a neural network, trained to recognize pixel with skin characteristics, is used to completely localize the face by delimiting the region surrounding the vertical symmetry axis which colour characteristics agree with skin ones according to TSL colour space. At this stage, we explore three strategies to perform region delimitation. Finally, a quantitative measurement criterion is used to record localization quality. Experiments of the proposed method were carried out on images of XM2VTS database and also on a set of non-standard images.

Various diseases caused by pathogenic parasites and fungi may be characterized by shape based structures. No significant attempt has been made so far to categorize such parasites by their shape properties, which can make the task of information retrieval much easier than annotating all of them separately. Here we present an automatic classification system which can retrieve the parasite or fungi’s information from the database using shape based information. To reduce time complexity of the information retrieval parasites having more or less identical shapes are clustered in the same group. A set of shape descriptors, generated by trace transform has been used to characterize structure of worms. Backpropagation neural network is trained, which leads to 85.71% accuracy of classification using statistically significant shape features.

The exploitation of SFCs in image compression could constitute an attractive alternative for efficient image coder design. Different ways of SFC exploitation have been explored in the literature. In this paper, we present a comparative study of different SFC based compression methods. This study includes the specification of new scan or SFC evaluation metrics, the presentation of different methods basics, the evaluation, classification, modeling of existing method and the analysis of their potential usage in image compression.

In this paper it is proposed a boundary based classifier that is inspired by SVDD and makes an important role for gravity center of training samples. In the proposed method all training samples intervene in determining the classifier boundary. Consequently, the relevant classifier isn’t placed in the group of the support vector machines. Due to the employment of this idea, this method is called "Quasi Support Vector Data Description (QSVDD)". The ability of this method to eliminate the effect of noisy training samples on synthetic data is shown. Experiments on real data sets show that the proposed method describes more accurately lots of real data sets than SVDD.

Noise has been a primary deterrent in signal transmission which results in faulty information after signal processing, reducing their usability. It is important that signals should be free from unwanted, random variations so that the errors caused due to faulty representation of the original signal can be minimized. Digital filter describes a linear time invariant system used to perform spectral shaping or frequency selective filtering. Design of digital filter using averaging technique is a simple type of FIR filter without assigning any value of cut off frequency. In our work, Simple Moving Average (SMA) and Exponentially Weighted Moving Average (EWMA) based signal filtering techniques are applied on corrupted signal, having different signal to noise ratio (SNR, to filter out noise from a discretely sampled signal. Performance evaluation of these techniques has been carried out using DSP Starter Kit (DSK) having processor TMS320C6713 with Code Composer Studio and is verified through simulation using MATLAB. The algorithm of EWMA technique is synthesized on FPGA platform to estimate resource utilization and timing constraints. The motivation for this research work was derived from the quest to develop a FPGA based simple filtering technique as compared to the conventional FIR filters to reduce design cycle of the system.

Moving object detection has been widely used in diverse discipline such as intelligent transportation systems, airport security systems, video monitoring systems, and so on. In this paper, we propose an efficient moving object detection method using enhanced edge localization mechanism and gradient directional masking for video surveillance system. In our proposed method, gradient map images are initially generated from the input and background images using a gradient operator. The gradient difference map is then calculated from gradient map images. The moving object is then detected by using appropriate directional masking and thresholding. Simulation results indicate that the proposed method consistently performs well under different illumination conditions including indoor, outdoor, sunny, and foggy cases. Moreover, it outperforms well known edge based method in terms of detecting moving objects and error rate. Moreover, the proposed method is computationally faster and it is applicable for detecting moving object in real-time.

Video tracking is one of the most active research topics recently. Tracking of objects and humans has a very wide set of applications such as teleconferencing, surveillance, and security. We propose a new tracker to enhance the tracking process by making use of SURF descriptor and Particle filter. SURF is one of the fastest descriptors which generates a set of interesting points which are invariant to various image deformations (scaling, rotation, illumination) and robust against occlusion conditions during tracking. Particle filter is one of the commonly used methods in video tracking to solve non-linear and non-Gaussian problems. Particle filter generates a random set of points called particles or samples for any target to be used for tracking through the process of the algorithm. But the fact that the initial particles are chosen randomly causes degradation in efficiency and reliability of the tracking process. It is possible to lose the tracked target at any frame if any change happened in the scene. Previous researches proposed an integration of Particle algorithm and scale invariant feature transform (SIFT) descriptor to overcome potential problems. SIFT is a predecessor of SURF and shares the same characteristics except that SURF is much faster. A comparative study was held between the traditional particle filter, SIFT-Particle tracker and the proposed tracker. The proposed SURF-Particle tracker proved to be more efficient, reliable and accurate than traditional particle filter and SIFT-Particle tracker. The idea of the proposed tracker is to use the discriminative interest points generated by the SURF descriptor as the initial particles/ samples to be fed into particle filter instead of choosing these particles randomly as done in traditional simple particle filter. Experimental results using the Actions as Space-Time Shapes Dataset of the Weizmann Institute of Science proved the correctness of the proposed idea and showed improved efficiency and accuracy resulted from using our proposed tracker over traditional simple particle filter and SIFT-Particle tracker. It also proved to be faster than SIFT-Particle.

Reducing the Radiation Dose in Multi Slice Computerized Tomography MSCT/CT is a significant concern. The Non-Linear Bilateral Filter BF was proved to have the property of de-noising digital images without jeopardizing the fine structures. This paper tests the BF performance on low dose CT by using Image Texture Metrics which have not been reported in literature. Set of CT images of dedicated CT phantom were acquired at four different radiation doses by means of minimizing the X-Ray Tube Current. As radiation dose is lowered, the noise will unavoidably increase degrading the diagnostic value of the CT image. The BF was applied to achieve image space noise removal. The value of each BF parameter was changed set of times. The quantitative assessment of the amount of noise reduction was done using eight metrics based on image texture descriptors that have not been tried before. Particularly, we used three histogram moments (Variance, Skewness, Kurtosis) and five co-occurrence matrix descriptors (Correlation, Contrast, Uniformity, Homogeneity, Entropy). The results showed that these descriptors are reliable metrics to assess BF performance. Each image descriptor value -after applying BF on low dose CT images- is enhanced toward the full dose CT image. Therefore, these metrics have provided additional proofs about the capability of BF toward enhancing the diagnostic value of the low dose MSCT. We concluded that: 1-) Texture descriptors are reliable measures similar to other metrics that are commonly used in literature, and 2-) BF can contribute to reduce X-Ray dose in routine CT. Also, the results have leaded to propose the effective procedure to employ BF on CT.

In this paper, a novel image projection technique for face recognition application is proposed which is based on linear discriminant analysis (LDA) coined Kernel Weighted Scatter Two Dimensional Discriminant Analysis (KWS2DDA). The projection is performed through 2-direction which simultaneously works in row and column directions to solve the small sample size problem. This nonlinear dimensionality reduction algorithm has several interesting characteristics. It’s overcomes the singularity problem, while achieving efficiency. In order to improve the performance of the proposed algorithm, we introduce Gaussian RBF kernel functions. We have performed multiple face recognition experiments to compare KWS2DDA with other dimensionality reduction methods showing that KWS2DDA consistently gives the best result than the other method.

The computer vision techniques are required for the identification of leafy vegetables for the development of veggie vision applications. This paper presents the mean and range color features based identification of leafy vegetables. Initially, a total of 18 RGB and HSI color features are chosen. A reverse engineering process is adopted for reduction of features. Finally 12 mean and range features of RGB and HSI color features are selected based on the performance. A BPNN based classifier is used for identification of vegetables. The identification rate is in the range 92-100% for ten types of vegetables. The work finds applications in automatic vending, packing and grading of vegetables, food preparation and the like.

A novel algorithm for 2D object orientation and scaling factor estimation, is proposed in this paper. The proposed method is accurate, effective, computationally efficient and fully- automated. The object orientation is calculated by using object principal axes estimation. The main contribution of the proposed approach is the utilization of a 2D empirical mode like decomposition (EMD-like) as a new workspace for principal axes and scaling determination. The EMD algorithm can decompose any nonlinear and non-stationary data into a number of intrinsic mode functions (IMFs). When the object is decomposed by empirical mode like decomposition (EMD-like), the IMFs of the object, provide a workspace with very good properties for calculating the principal axes. The method was evaluated on synthetic and real images. The experimental results demonstrate the effectiveness and the accuracy of the method, both in orientation and scaling estimations.

An efficient implementation of discrete cosine transform (DCT) computations is presented based on the Ramanujan ordered number DCT (RDCT), a fast multiplierless DCT algorithm. Due to the simple form of the factorized matrices, the derived architecture can be easily constructed from the cascade of only two types of parameterized hardware modules: shifters and adders. The proposed implementations have many features and advantages, including low complexity, high-throughput and regularity. The regularity of RDCT algorithm and careful operation scheduling has resulted in a very efficient implementation of a multiplierless RDCT in Xilinx Spartan3 FPGA in the terms of logic requirements.

Semantic similarity between word senses is hot topic in many applications of computational linguistics and artificial intelligence, such as word sense disambiguation, information extraction, semantic annotation and ontology learning. Many methods for calculating word sense similarity have been proposed. In recent years the methods based on WordNet have shown its talents and attracted great concern. In the paper, we present a new method in WordNet for calculating word sense similarity, which is noun and is-a relation based. We evaluate our method on the data set of Rubenstein and Goodenough, which is traditional and widely used. The correlation with human judgment is o.8804 in proposed measure, which is more close to human judgments than related works. Experiments show that our new measure significantly outperformed than other existing computational methods.

Video can be regarded as three dimensional spatio-temporal volume, in which human ac-tion is a three dimensional shape (3D shape) surrounded by the spatio-temporal silhouette surface. The type of human action depends on the shape of the silhouette surface. In this pa-per, we proposed a new feature called Oriented Gradient Histogram of Slide Blocks by build-ing dense overlapping spatio-temporal slide blocks to detect the shape of the 3D silhouette surface of the human action. Sparse coding is adopted to represent videos based on the new feature and Random Forest is utilized to classify the types of human actions. Experiments on KTH and Weizmann human action datasets demonstrate that the new feature can describe the spatio-temporal silhouette surface correctly, accordingly recognize the human action types accurately.