Vahid Abolghasemi

An edge-based color-aided method for license plate detection

In this paper, the problem of license plate detection is considered. Low quality images due to severe illumination conditions, vehicle motion, viewpoint and distance changes, complex background, etc. are some of popular problems which have to be considered. In order to alleviate these problems, two different image enhancement methods (using intensity variance and edge density) are proposed. The aim is to increase contrast of plate-like regions to avoid missing plate location especially in poor quality images. Furthermore, a novel match filter is designed to detect candidate regions as plate. This filter models the vertical edge density of plate region regarding its neighborhood. As the filtering procedure is simple, this approach can be used for real-time applications. In the proposed method, we also use colored texture in the plate as a cue for plate detection. This feature is preserved under viewpoint change. In order to characterize the color information in plate, the MNS (multimodal neighborhood signature) method is used. A well-organized database, consisting of car images with different known distances and viewing angels have been prepared to verify the performance of plate detection algorithm. This database can be used to establish a precise evaluation of the proposed method and any other related work. The results of experiments on different type of car images in complex scenes confirm the robustness of proposed method against severe imaging conditions.

Fast and incoherent dictionary learning algorithms with application to fMRI

In this paper, the problem of dictionary learning and its analogy to source separation is addressed. First, we extend the well-known method of K-SVD to incoherent K-SVD, to enforce the algorithm to achieve an incoherent dictionary. Second, a fast dictionary learning algorithm based on steepest descent method is proposed. The main advantage of this method is high speed since both coefficients and dictionary elements are updated simultaneously rather than column-by-column. Finally, we apply the proposed methods to both synthetic and real functional magnetic resonance imaging data for the detection of activated regions in the brain. The results of our experiments confirm the effectiveness of the proposed ideas. In addition, we compare the quality of results and empirically prove the superiority of the proposed dictionary learning methods over the conventional algorithms.

A robust approach for optimization of the measurement matrix in Compressed Sensing

In this paper we address the problem of measurement matrix optimization in Compressed Sensing (CS) framework. Although the measurement matrix is generally selected randomly, some methods have been recently proposed to optimize it. It is shown that the optimized matrices can improve the quality of reconstruction and satisfy the required conditions for an efficient sampling. We propose a new optimization method with the aim of decreasing the “Mutual Coherence”. Defining a new cost function, we suggest to use a Gradient descent algorithm for this optimization problem. The advantages are less computational complexity, which makes the method suitable for large-scale problems, more robustness, and higher incoherence between the measurement matrix and sparsifying matrix (dictionary). By conducting several experiments, we obtained promising results which confirm a considerable improvement compared to those achieved by other methods.

Simultaneous denoising and interpolation of 2D seismic data using data-driven non-negative dictionary learning

VNMF combines dictionary learning and sparse coding to find atoms of basis matrix.Using non-negativity constraint to induce sparsity and reduce the solution space.Utilizing all patches of the data to learn a dictionary for precise performance.The algorithm uses lower number of atoms in learning.Simultaneous denoising and interpolation using one minimization problem. As a major concern, the existence of unwanted energy and missing traces in seismic data acquisition can degrade interpretation of such data after processing. Instead of analytical dictionaries, data-driven dictionary learning (DDL) methods as a flexible framework for sparse representation, are dedicated to the problem of denoising and interpolation. Due to their meaningful geometric repetitive structures, seismic data are intrinsically low-rank in the time-space domain. On the other hand, noise and missing traces increase the rank of the noisy data. Therefore, the clean data, unlike noise and missing traces, can be modeled as a linear combination of a few elements from a learned dictionary. In this paper, a parts-based 2D DDL scheme is introduced and evaluated for simultaneous denoising and interpolation of seismic data. A special case of versatile non-negative matrix factorization (VNMF) is used to learn a dictionary. In VNMF, smoothness constraint can improve interpolation, and sparse coding helps improving denoising. The proposed method is tested on synthetic and real-field seismic data for simultaneous denoising and interpolation. Through experimental results, the proposed method is determined to be an effective and robust tool that preserves significant components of the signal. Comparison with four state-of-the-art methods further verifies its superior performance.

A fast algorithm for license plate detection

In this paper we propose a method for detection of the car license plates in 2D gray images. In this method we first estimate the density of vertical edges in the image. The regions with high density vertical edges are good candidates for license plates. In order to filter out clutter regions possessing similar feature in the edge density image, we design a match filter which models the license plate pattern. By applying the proposed filter on the edge density image followed by a thresholding procedure, the locations of license plate candidates are detected. We finally extract the boundary of license plate(s) using the morphological operations. The result of experiments on car images (taken under different imaging conditions especially complex scenes) confirms the ability of the method for license plate detection. As the complexity of the proposed algorithm is low, it is considerably fast.

Removing Ballistocardiogram Artifact From EEG Using Short- and Long-Term Linear Predictor

In this paper, a novel source extraction method is proposed for removing ballistocardiogram (BCG) artifact from EEG. BCG appears in EEG signals recorded simultaneously with functional magnetic resonance imaging. The proposed method is a semiblind source extraction algorithm based on linear prediction technique. We define a cost function according to a joint short- and long-term prediction strategy to extract the BCG sources. We call this method SLTP-BSE standing for short- and long-term prediction blind source extraction. The objective of this work is to 1) model the temporal structure of the sources using short-term prediction and 2) impose the prior information about the BCG sources using long-term prediction. These two procedures are simultaneously implemented to optimize the system. The performance of the proposed method is evaluated using both synthetic and real EEG data. The obtained results show that the proposed technique is able to remove the BCG artifact while preserving the task-related parts of the signal. The results of SLTP-BSE are compared with those of well-known BCG removal techniques confirming the superiority of the proposed method.

A constrained NMF algorithm for bold detection in fMRI

In this paper the application of Nonnegative Matrix Factorization (NMF) to Functional Magnetic Resonance Images (fMRIs) is addressed. We attempt to blindly separate the sources of fMRI mixtures. However, our interest is to find only one particular source (task-related source), which indicates the active area in the brain. We utilize the prior knowledge about time course of the corresponding source to automatically extract it. By proposing a template based on this prior knowledge we set up a constrained local cost function which is to be minimized. In order to be able to achieve such an optimization, the Hierarchical Alternate Least Square (HALS) algorithm is adopted. The advantage of the proposed method is to simultaneously separate and distinguish the source of interest from other sources.

Heart and lung sound separation using periodic source extraction method

In this paper a novel algorithm for blind source extraction of quasi-periodic signals is developed. It is assumed that the cycle frequencies of the sources are known a priori. Necessary and sufficient conditions are introduced and Jacobi method for diagonalization of complex matrices is used to find the solution. A warping method is also introduced to adjust the cycle frequencies in the signals with time varying periods. The proposed algorithm is applied to simulated data and mixtures of heart and lung sounds, and effectiveness and performance of the algorithm are verified.

Detecting license plate using texture and color information

We consider the problem of detecting car license plate in 2D images. In the first stage of algorithm we enhance the car image at interest regions by estimating image edge density. The enhanced image is then processed using the proposed match filter to extract candidates for car plate. The simplicity of this process makes it suitable for real-time applications. In order to extract car plate among the candidate regions we consider two constraints: geometrical criterion and colour descriptors. The first criterion simply involves geometrical constraints to filter out many of false positives from candidate list. In the second stage of filtering we use colour information on plate to detect license plate among candidate regions. For this purpose we apply multimodal neighbourhood signature algorithm to model license plate. The result of experiments on car images in real complex scene confirms the robustness of the proposed method against severe imaging condition.

Persian signature verification using improved Dynamic Time Warping-based segmentation and Multivariate Autoregressive modeling

In this paper, we present an online signature verification system based on Dynamic Time Warping (DTW)-based segmentation technique combined with Multivariate Autoregressive (MVAR) modeling. We also use multilayer perceptron neural network architecture as data classifier. The input data that has been used is (xj,yj) coordinates of signatures drawn from a Persian database. We compare two different DTW algorithms in terms of their effect in improving the alignment between the signature sample and a master signature reference for the subject writer. Our database includes 1250 genuine signatures and 750 forgery signatures that were collected from a population of 50 human subjects. We used 75% of samples for training and 25% for testing. We achieved an accuracy of 88.8% for a skilled forgery test which is a very promising result.