Earticle

Unimodal biometric systems have to contend with various inherent limitations, such as restricted degrees of freedom, non-universality, susceptibility to spoofing attacks and unacceptable error rates. Multibiometric systems, which fuse two or more biometrics traits together, are able to effectively overcome most of these problems. In this paper, different face traits are fused considering convenient acquiring of visible face and the intrinsic anti-spoofing of thermal face. Initially, the complex fusion strategies at both pixel level and feature level are proposed, which can provide higher discrimination superiority. The 2D-classification methods, including 2DPCA, 2DLDA, (2D)2PCA, (2D)2LDA and (2D)2FPCA are applied into the complex fusion, which can overcome the small size sample problems. Both identification and verification experiments are conducted on the NVIE visible and thermal face database. Various tests based on this database ascertain the efficacy of the proposed approaches in identification and verification. The better performances are in favor of the proposed approaches, FC_(2D)2LDA and FC_(2D)2FPCA, the training number 6 and 8, and the visible face fusion weight 0.4 and 0.6.

Majority of the videos that have been captured by mobile cameras are suffering from low quality due to either low end manufacturing designs or complicated operating environments and untrained users. Thus videos taken by hand held mobile cameras tend to suffer from different undesired slow motions that cause annoying shaky motion and jitter. It is desirable to stabilize the video sequence by removing the undesired motion between the successive frames. Current methods are applicable to only specific camera motion models; hence having limitation to process gorse motion. In this paper an efficient video stabilization algorithm for hand held camera videos has been proposed. The proposed algorithm uses differential global motion estimation with Taylor series expansion to improve the estimation efficiency. Affine motion model has been assumed to define the inter-frame error between consecutive frames. Motion vectors have been estimated analytically by solving the derivatives of the inter-frame error. After motion estimation Gaussian kernel filtering has been used to smoothen out estimated motion parameters. Inverse rotation smoothening has been applied to remove the rotation effect from the smoothed transformation chain. This has led to reduced accumulation error and minimizes the missing image area significantly. The performance of the proposed algorithm has been tested on real time videos and compared with existing algorithm.

In recent years, semi-supervised learning has been a hot research topic in machine learn-ing area. Different from traditional supervised learning which learns only from labeled data; semi-supervised learning makes use of both labeled and unlabeled data for learning purpose. Co-training is a popular semi-supervised learning algorithm which assumes that each exam-ple is represented by two or more redundantly sufficient sets of features (views) and addi-tionally these views are independent given the class. To improve the performance and ap-plicability of co-training, ensemble learning, such as bagging and random subspace has been used along with co-training. In this work, we propose to use the rough set based ensem-ble learning method with co-training algorithm (RSCO). Inherited the inherent characteris-tics of rough set, ensemble learning is expected to meet both the diversity and accuracy re-quirement. Finally experimental results on the UCI data sets demonstrate the promising per-formance of RSCO.

In this paper, the execution of some features based on wavelet transform are evaluated through classification of audio to speech and music using the MLP classifiers Optimized by Genetic Algorithm. Classification results show the wavelet features are completely successful in speech/music classification. Experimental comparisons using different wavelets are presented and discussed. By using some wavelet features, extracted from 1-second segments of the signal, we obtained 96.49% accuracy in the audio classification of the MLP classifiers optimized by genetic algorithm.

Automatic language identification plays an essential role in wide range of multi-lingual services. Automatic translators to certain language or routing an incoming telephone call to a human switchboard operator fluent in the corresponding language are examples of these applications that require automatic language identification. This paper investigates the usage of Linear Predictive Coding (LPC) and/or Mel Frequency Cepstral Coefficients (MFCC) with Artificial Neural Network (ANN) for automatic language identification. Different orders for the LPC and MFCC have been tested. In addition, different hidden layers, different neurons in every hidden layers and different transfer functions have been tested in the ANN. Three languages; Arabic, English and French have been used in this paper to evaluate the performance of the automatic language identification systems.

In this paper, an enhanced thermal face recognition method, namely GMFB, is proposed. Initially, Gabor Jet Descriptor (GJD) is extracted from each thermal image with five scales and eight orientations. Then, the Modified Fisher (MF) criterion is implemented on the feature vector for every scale. Finally, the Borda count (BC) matching method is used to get higher matching score. Our proposed method enhances the discrimination ability of the feature vector significantly. Experiments conducted on NVIE thermal face database show that the proposed approach outperforms the state-of-the-art methods.

Texture analysis such as segmentation and classification plays a vital role in computer vision and pattern recognition and is widely applied to many areas such as industrial automation, bio-medical image processing and remote sensing. Over the last decade, several studies on texture analysis propose to model texture as a probabilistic process that generates small texture patches. In these studies, texture is represented by means of a frequency histogram that measures how often texture patches from a codebook occur in the texture. In the codebook, the texture patches are represented by a collection of filter bank responses. The resulting representations are called textons. A recent study claims that textons based on gray values outperform textons based on filter responses. Textons refer to fundamental micro structures in natural images and are considered as the atoms of pre-attentive human visual perception. This paper describes a novel technique of image segmentation for texture images based on six different texton patterns and morphological transforms.

The character segmentation technology for printed documents plays an important role in optical character recognition, ticket information identification, postal code identification, automatic license plate recognition and so on. In this paper, a Chinese characters segmentation algorithm for complicated printed documents is proposed for the application in paper watermarking system. In this application, the algorithm aims to achieve high accuracy Chinese character segmentation and high consistent segmentation between the digital version images and print-scanned version images for the same documents. In this method, three main steps are included: connected regions recognition, connected regions merging, and fine-gained segmentation. Experiments show the effectiveness of the proposed algorithm.

In this paper, we propose a novel and effective image model—Image Latent Semantic Analysis (ILSA) for extracting latent semantic features of face image, and recognizing face with Support Vector Machine (SVM). The novel feature extraction by the ILSA model can be better overcome the impact of some negative factors, such as the image quality fuzzy, illumination changes effect. The main contribution of the paper is that the ILSA features can obtain a wealth of information than the conventional image semantic features and has a stronger expression and classification abilities than the low-level features. The experimental results on the ORL and large-scale FERET databases show that the proposed algorithm significantly outperforms other well-known algorithms.

Three dimensional (3D) reconstruction of the tumor from medical images is an important operation in the medical field as it helps the radiologist in the diagnosis, surgical planning and biological research. Thus in this paper, we propose an effective and efficient approach to 3D reconstruction of brain tumor and estimation of its volume from a set of two dimensional (2D) cross sectional magnetic resonance (MR) images of the brain. In the first step, MR images are preprocessed to improve the quality of the image. Next, abnormal slices are identified based on histogram analysis and tumor on those slices is segmented using modified fuzzy c-means (MFCM) clustering algorithm. Next, the proposed enhanced shape based interpolation technique is applied to estimate the missing slices accurately and efficiently. Then, the surface mesh of the tumor is reconstructed by applying the marching cubes (MC) algorithm on a set of abnormal slices. The large number of triangles generated by the MC algorithm was reduced by our proposed mesh simplification algorithm to accelerate the rendering phase. Finally, rendering was performed by applying Phong lighting and shading model on the reconstructed mesh to add realism to the 3D model of the tumor. The volume of the tumor was also computed to assist the radiologist in estimating the stage of the cancer. All experiments were carried out on MR image datasets of brain tumor patients and satisfactory results were achieved. Thus, our proposed method can be incorporated into the computer aided diagnosis (CAD) system to assist the radiologist in finding the tumor location, volume and 3D information.

Block matching for motion estimation has been widely used in video compression for effi-cient transmission and storage of video bit stream by reducing the temporal redundancy ex-isting in a video sequence. The motion estimation is a process to predict the motion between two successive frames. This paper is primarily a review of the block matching algorithms using fast computational and winner-update strategies. The paper describes and analyses different types of block matching algorithms, namely Full Search (FS), Fast Computational of Full Search (FCFS), Three Step Search (TSS), New Three Step Search (NTSS), Three Step Search with Winner Update Strategy (WinUpTSS), Four Step Search (FSS) and Diamond Search (DS) algorithms.

The information extraction process of medical image, for example heart image from specific camera, is full of complexities and noises. As a result, cost spent on such processing like time and resources is high, especially for large and complex amount of information. This paper uses machine vision-based approach to address the challenges. This approach primarily includes four stages with different algorithms to deal with the medical heart images. Firstly, smoothing algorithm is used to reduce the noise. Secondly, filtering algorithm is used for image analysis so as to identify the target area. Thirdly, further enhancement algorithm is used to figure out the image features within the target area, finding out the basic outline of the image. Eventually, reduction algorithm is utilized to convert the original images into more smooth and precise pictures. Experimental results show that machine vision-based medical image processing algorithm can accurately extract the relevant data and achieve better results, comparing with the traditional image processing method.