Thai Hoang Le

Fingerprint reference point detection for image retrieval based on symmetry and variation

Reference point plays an important role in fingerprint identification systems. The reference point is widely used for the fingerprint retrieval in large-scale databases. This paper proposes a novel algorithm for detecting a convex core point as a unique reference point consistently and accurately for all types of fingerprints. In order to detect robust core point candidates, a modified complex filter, called semi-radial symmetry filter, is proposed to detect correctly rotational symmetries of core points. Moreover, a vertical orientation variation feature, called VORIV feature, is proposed to remove spurious core points and concave core points. Therefore, the proposed technique computes the Variation and Symmetry Combined Energy (VSCOME). Then, the reference point is located by searching the global VSCOME maximum. The experimental results on the public database (FVC2004 DB1 set A) show that the proposed technique exhibits a very high robustness and gets the best performance in comparing with other approaches in literature.

Palmprint verification using GridPCA for Gabor features

Palmprint recognition is a challenging problem for reliable person authentication due to the complex patterns of palmprint image. In this paper, we propose a novel algorithm for palmprint verification based on using Grid-PCA for Gabor features. Our proposed algorithm includes the following steps: (1) Gabor feature extraction by using a Gabor function of multiple scales and orientations with the aim that optimize local properties in both spatial and frequency domain. (2) Grid-PCA is applied to effectively reduce the feature space dimension of each Gabor-filtered image. The experimental results for the verification on public palmprint databases of Indian Institute of Technology Delhi and Hong Kong Polytechnic University show that the proposed method exhibits a very high robustness and gets the best performance in comparing to other approaches in literature.

Keystroke dynamics extraction by independent component analysis and bio-matrix for user authentication

Keystroke dynamics is unique specific characteristics used for user authentication problem. There are many researches to detect personal keystroke dynamics and authenticate user based on these characteristics. Most researches study on either the key press durations and multiple key latencies (typing time) or key-pressed forces (pressure-based typing) to find the owned personal motif (unique specific characteristic). This paper approaches to extract keystroke dynamics by using independent component analysis (ICA) through a standardized bio-matrix from typing sound signals which contain both typing time and typing force information. The ICA representation of keystroke dynamics is effective for authenticating user in our experiments. The experimental results show that the proposed keystroke dynamics extraction solution is feasible and reliable to solve user authentication problem with false acceptance rate (FAR) 4.12% and false rejection rate (FRR) 5.55%.

On approaching 2D-FPCA technique to improve image representation in frequency domain

A novel approach based on structure information extraction in frequency domain is proposed for image representation problem. Regarding this problem, a new subspace method based on Two-dimensional Fractional Principle Component Analysis (2D-FPCA) in frequency domain is applied to images, thus extracting the texture information. In order to extract the structure information, the system utilizes this new subspace as the bilateral consideration of 2D-FPCA technique called B2D-FPCA. For this purpose: (1) we first introduce the theory of 2D-FPCA based on the definition of fractional variance and fractional covariance matrix; (2) then show its improvement called Bilateral 2D-FPCA and (3) the robustness of 2D-DCT is also described as the preprocessing step. This approach is applied to facial expression representation problem to prove the stability and robustness of the proposed framework. For demonstration, facial expressions datasets (JAFFE, Pain expression subset and Cohn-Kanade) are used in order to compare the proposed framework with some other approaches.

Combining global features and local minutiae descriptors in genetic algorithms for fingerprint matching

Fingerprint matching is an important and challenging problem in fingerprint recognition. Many approaches have been proposed for fingerprint matching such as minutiae point pattern-based techniques, orientation pattern-based techniques, ridge-based techniques, global and local features combination-based techniques (GLF-BCT). In recent research, GLF-BCT methods achieved good performance even when a large portion of fingerprints in the database are of poor quality. In this paper, we would like to improve the GLF-BCT model using Genetic Algorithm (GA) that we aim to achieve higher efficiency in fingerprint recognition. In detail, the proposed model is a combination of the advantage of local minutiae descriptors (ability of increasing the distinctiveness degree between two different fingerprint images) with the advantage of the global features (identifying the optimal or near optimal global alignment between two fingerprints) to improve the reliability of GA fitness assignment in fingerprint matching. This method is called the Fingerprint Matching based on Combining Global features and Local minutiae Descriptors in Genetic Algorithms (FM-CGLD-GA). The experimental results on the FVC2004 database show the effectiveness and superiority of the proposed method in comparing to other approaches.

A hybrid approach of AdaBoost and Artificial Neural Network for detecting human faces

The human face image recognition is one of the prominent problems at present. Recognizing human faces correctly will aid some fields such as national defense and person verification. One of the most vital processing of recognizing face images is to detect human faces in the images. Some approaches have been used to detect human faces. However, they still have some limitations. In the paper, we will consider some popular methods, AdaBoost, artificial neural network (ANN) etc., for detecting human faces. Then we will propose a hybrid model of combining AdaBoost and artificial neural network to solve the process efficiently. The system which was build from the proposed model has been conducted on database CalTech. The recognition correctness is more than 96%. It shows the feasibility of the proposed model.

GridLDA of Gabor wavelet features for palmprint identification

In this paper, we propose a novel palmprint recognition algorithm based on using GridLDA for Gabor wavelet features. Our proposed method includes two main steps for palmprint feature extraction: (1) Local invariant features are extracted by computing the Gabor wavelet Engergy of the original images that handles the palm structure and the variations of illumination. (2) An improved two-dimensional Linear Discriminant Analysis, called GridLDA, is then applied to further remove redundant information and form a discriminant representation more suitable for palmprint recognition. The experimental results for the identification on public database of Hong Kong Polytechnic University (PolyU) demonstrate the effectiveness of the proposed method.

A fuzzy neural network for Vietnamese character recognition

In this research, on the background of a combination between Neural Network and Fuzzy Logic theories applying to the field of recognition, we have developed our four-layer feedforward Fuzzy Neural Network (FNN) as a solution to solve Vietnamese character recognition problems. Our first-step experiments of FNN were applied for 28 Vietnamese accent characters in VNI-Times font. Our obtained testing results prove the availability of FNN as a good solution even to bad-quality patterns with high distortion rate.

Discriminant Orientation Feature for Palmprint Recognition

In this paper, we propose a novel feature for palmprint recognition, called Discriminant Orientation Feature (DORIF) based on using Modified Finite Radon Transform (MFRAT) and Two Directional Two Dimensional Linear Discriminant Analysis (2D)2LDA. Our proposed method includes two main steps for palmprint feature extraction: (1) Local invariant orientation features are extracted by using MFRAT that handles the palm structure and the variations of illumination and rotation. (2) (2D)2LDA is then applied to further remove redundant information and form a discriminant representation more suitable for palmprint recognition. The experimental results for the identification on public database of Hong Kong Polytechnic University (PolyU) demonstrate the effectiveness of the proposed method.

Deep generic features and SVM for facial expression recognition

Motivated by the newly recent trend in pattern recognition - convolutional neural network (CNN), we introduce a new fusion method based on CNN and support vector machines (SVM) for facial expression recognition problem. Our study puts the deep generic features from CNN and SVM together which is more efficient than CNN only. We investigate our proposed method on Cohn-Kanade dataset and achieve 96.04% in accuracy rate which is better than other state-of-the-art methods.