Yanlai Li

Palmprint recognition based on translation invariant Zernike moments and modular neural network

This paper introduces a new approach, TIZMs & MNN, for palmprint recognition. It uses translation invariant Zernike moments (TIZMs) as palm features, and a modular neural network (MNN) as classifier. Translation invariance is added to the general Zernike moments which have very good property of rotation invariance. A fast algorithm for computing the TIZMs is adopted to improve the computation speed. The pattern set is set up by eightorder TIZMs. Because palmprint recognition is a large-scale multi-class task, it is quite difficult for a single multilayer perceptrons to be competent. A modular neural network is presented to act the classifier, which can decompose the palmprint recognition task into a series of smaller and simpler two-class subproblems. Simulations have been done on the Polyu_PalmprintDB database. Experimental results demonstrate that higher identification rate and recognition rate are achieved by the proposed method in contrast with the straight-line segments (SLS) based method [2].

Adaptive weighted fusion of local kernel classifiers for effective pattern classification

The theoretical and practical virtual of local learning algorithms had been verified by the machine learning community. The selection of the proper local classifier, however, remains a challenging problem. Rather than selecting one single local classifier, in this paper, we propose to choose several local classifiers and use adaptive fusion strategy to alleviate the choice problem of the proper local classifier. Based on the fast and scalable local kernel support vector machine (FaLK-SVM), we adopt the self-adaptive weighting fusion method for combining local support vector machine classifiers (FaLK-SVMa), and provide two fusion methods, distance-based weighting (FaLK-SVMad) and rank-based weighting methods (FaLK-SVMar). Experimental results on fourteen UCI datasets and three large scale datasets show that FaLK-SVMa can chieve higher classification accuracy than FaLK-SVM.

Gaussian ERP Kernel Classifier for Pulse Waveforms Classification

While advances in sensor and signal processing techniques have provided effective tools for quantitative research on traditional Chinese pulse diagnosis (TCPD), the automatic classification of pulse waveforms is remained a difficult problem. To address this issue, this paper proposed a novel edit distance with real penalty (ERP)-based k-nearest neighbors (KNN) classifier by referring to recent progresses in time series matching and KNN classifier. Taking advantage of the metric property of ERP, we first develop a Gaussian ERP kernel, and then embed it into kernel difference-weighted KNN classifier. The proposed Gaussian ERP kernel classifier is evaluated on a dataset which includes 2470 pulse waveforms. Experimental results show that the proposed classifier is much more accurate than several other pulse waveform classification approaches.

Step acceleration based training algorithm for feedforward neural networks

This paper presents a very fast step acceleration based training algorithm (SATA) for multilayer feedforward neural network training. The most outstanding virtue of this algorithm is that it does not need to calculate the gradient of the target function. In each iteration step, the computation only concentrates on the corresponding varied part. The proposed algorithm has attributes in simplicity, flexibility and feasibility, as well as high speed of convergence. Compared with the other methods, including the conventional backpropagation (BP), conjugate gradient, and weight extrapolation based BP, many simulations confirmed the superiority of this algorithm in terms of converging speed and computation time required.

Tongue coating image retrieval

Along with the rapid growth of medical data, image retrieval, a kind of technology for browsing, searching and retrieving similar images of the given image, has become increasingly important from a large database of digital images. Tongue coating is the most important characteristic to reveal the pathological changes of the tongues for identifying diseases. In this paper, an efficient and effective technique is proposed to retrieve coating images. We obtain the pixel template value of pixels by applying thresholding segmentation based relative entropy. Then we use a Reduced K Nearest Neighbor algorithm to extract 20-dimension feature vector based on a prior layout distribution. Finally, a distance based the cumulative ratio is proposed for tongue coating image matching. The experimental results indicate that the proposed scheme eliminates the imprecision and uncertainty associated with medical tongue coating analysis.

Modular neural network structure with fast training/recognition algorithm for pattern recognition

In this paper, modular neural network structure with fast training/recognition algorithm for pattern recognition task decomposition is presented. After the modular neural network is described, a new training algorithm, named non-gradient (NG) training algorithm, is proposed to train the sub-modules. The inputs error of the output layer is taken into account. Four classes of solution equations for parameters are deducted respectively. The advantage of the presented algorithm is that it doesnpsilat need calculating the gradient of error function at all. In each iteration step, the weight or threshold can be optimized one by one with other parameters fixed. In the recognition stage, a new and fast JUMP recognition algorithm is proposed to save the recognition time. Effectiveness of the presented scheme is demonstrated by a palmprint recognition experiment.

Parameter by Parameter Algorithm for Multilayer Perceptrons

This paper presents a parameter by parameter (PBP) algorithm for speeding up the training of multilayer perceptrons (MLP). This new algorithm uses an approach similar to that of the layer by layer (LBL) algorithm, taking into account the input errors of the output layer and hidden layer. The proposed PBP algorithm, however, is not burdened by the need to calculate the gradient of the error function. In each iteration step, the weights or thresholds can be optimized directly one by one with other variables fixed. Four classes of solution equations for parameters of networks are deducted. The effectiveness of the PBP algorithm is demonstrated using two benchmarks. In comparisons with the BP algorithm with momentum (BPM) and the conventional LBL algorithms, PBP obtains faster convergences and better simulation performances.

Tongue color visualization for local pixel

Tongue diagnosis is a unique and important diagnostic method in Traditional Chinese Medicine (TCM). It is used to observe abnormal changes in the tongue color for identifying syndrome patterns. However, due to its qualitative, subjective and experience-based nature, it is hard to represent and visualize tongue color in computerized tongue diagnosis. This will undoubtedly limit the application of tongue color analysis in clinical practice. In this paper, we proposed a visualization system for local pixel separation to recognize this intrinsic pathological trait: tongue color. This system captures a digital image, extracts a contour of the tongue and constructs a lookup table. Finally, this lookup table is employed to construct some figures of different classes for the representation and visualization tongue color. The experimental results can draw a conclusion: Tongue color is a kind of pixel distribution.

A New Layer by Layer training algorithm for multilayer feedforward neural networks

A New Layer by Layer (NLBL) training algorithm for speeding up the training of multilayer feedforward neural networks is presented in this paper. It uses an approach similar to that of the Layer by Layer (LBL) algorithm, taking into account the input errors of the output layer and hidden layer. The proposed NLBL algorithm, however, is not burdened by the need to calculate the gradient of the error function. Furthermore, it has avoided the stalling problem exists in the LBL algorithm. In each iteration step, the weights or thresholds can be optimized directly one by one with other variables fixed. Four classes of solution equations for parameters of networks are deducted. In comparisons with the BP algorithm with momentum (BPM) and the conventional LBL algorithms, NLBL algorithm obtains faster convergences and better simulation performances when applied into a real world oil-gas prediction problem.

Tongue image texture segmentation based on gabor filter plus normalized cut

Texture information of tongue image is one of the most important pathological features utilized in practical Tongue Diagnosis because it can reveal the severeness and change tendency of the illness. A texture segmentation method based on Gabor filter plus Normalized Cut is proposed in this paper. This method synthesizes the information of location, color and texture feature to be the weight for Normalized Cut, thus can make satisfactroy segmentation according to texture of tongue image. The experiments show that the overall rate of correctness for this method exceeds 81%.