Ping Guo

Blind image restoration based on wavelet analysis

Blind image restoration is important and full of challenge as an issue of image processing. Many conventional approaches have been developed to restore the original image at present. But since there are two unknown things: the original image and the point spread function (PSF), in many case it is difficult to reach the expected results, especially when the image is heavily degraded. In this paper, blind image restoration approach based on wavelet analysis is proposed. The key of the approach is the wavelet analysis implemented before the restoration with conventional approaches. Wavelet analysis returns some useful information about the degraded image as well as the restored image, and the information can help further restoration. The experimental results prove that the approach is satisfactory.

Epileptic EEG Signal Classification with ANFIS Based on Harmony Search Method

In this paper, the Adaptive Neuro-Fuzzy Inference System (ANFIS) is used for the classification of the epileptic electroencephalogram (EEG) signals. The ANFIS combines the adaptation capability of the neural networks and the fuzzy logic-based qualitative approach together. A given input/output data set is deployed to construct a fuzzy inference system, whose membership function parameters are trained using a back propagation algorithm in combination with a least squares method. However, the training method sometimes may lead to local optima. We here propose a new strategy of hybrid training algorithm based on the fusion of the ANFIS and Harmony Search (HS), HS-ANFIS, which is adopted to tune all the parameters of the ANFIS. The validity of our method is verified by numerical experiments.

Improved PSO algorithm with harmony search for complicated function optimization problems

Improved particle swarm optimization algorithm with harmony search (IHPSO) is proposed in this paper. This algorithm takes particle swarm search direction estimation mechanism and harmony search (HS) approach to particle swarm optimization (PSO) algorithm, which increases the search capability of PSO algorithm considerably. The proposed algorithm initializes a new search with harmony pitch adjusting or random selection when PSO search direction is estimated incorrectly. This can provide further opportunities of finding better solutions for the particle swarm by guiding the entire particle swarm to promising new regions of the search space and accelerating the search. PSO, HPSO and IHPSO, as well as other advanced PSO procedures from the literature were compared on several benchmark test functions extensively. Statistical analyses of the experimental results indicate that the performance of IHPSO is better than the performance of PSO and HPSO.

Multi-Level Kernel Machine for Scene Image Classification

Recently, a new representation for recognizing instances and categories of scenes called spatial Principal component analysis of Census Transform histograms (PACT) has shown its excellent performance in the scene image classification task. PACT captures local structures of an image through the Census Transform (CT), meanwhile, large scale structures are captured by the strong correlation between neighboring CT values and the histogram. However, the original spatial PACT only simply concatenates all levels compact histograms together, and discards the difference between various levels. In order to improve this problem, we propose a multi-level kernel machine method, which computes a set of base kernels at each level of pyramid of PACT, and finds optimal weights for best fusing all these base kernels for scene recognition. Experiments on two popular benchmark datasets demonstrate that our proposed multi-level kernel machine method outperforms the spatial PACT on scene recognition. Besides, our method is easy to be implemented comparing with spatial PACT.

Mallat Fusion for Multi-Source Remote Sensing Classification

The fusion of multi-source remote sensing data is to offer improved accuracies in land cover classification. The conventional fusion methods such as HIS and PCA can not enhance information and simultaneously preserve high fidelity. Thus, the fused image is not preferable for classification. In this paper, the multi-source remote sensing data fusion based on Mallat algorithm for classification is proposed. The purpose of fusion is to create a new image that is more suitable for recognition. The topic focuses on the pyramid decomposition and choosing coefficients in the fusion process. The performance of proposed method is assessed by statistical methods and its effectiveness also testified by classification accuracies

Optimization of training samples with affinity propagation algorithm for multi-class SVM classification

This paper presents a novel optimization method of training samples with Affinity Propagation (AP) clustering algorithm for multi-class Support Vector Machine (SVM) classification problem. The method of optimizing training samples is based on region clustering with affinity propagation algorithm. Then the multi-class support vector machines are trained for natural image classification with AP optimized samples. The feature space constructed in this paper is a composition of combined histogram with color, texture and edge descriptor of images. Experimental results show that better classification accuracy can be obtained by using the proposed method.

Learning multiple pooling combination for image classification

Recently sparse coding with spatial pyramid matching method has shown its excellent performance in image classification. Inspired by this technique, we present an image classification approach by learning the optimal Multiple Pooling Combination strategy based on Non-Negative Sparse Coding (MPC-NNSC) in this paper. First, non-negative sparse coding with three different pooling methods as well as spatial pyramid matching method are utilized to encode local descriptors for image representation, respectively. Then a promising weight learning approach is employed to find a set of optimal weights for best fusing all these pooling methods in different scales. Lastly, support vector machine classifier with linear and histogram intersection kernel is employed for the final classification task. Experiments on two popular benchmark datasets are presented and they demonstrate the better performance of the proposed scheme compared to the state-of-the-art methods.

Experimental Comparison of Geometric, Arithmetic and Harmonic Means for EEG Event Related Potential Detection

In this paper, we experimentally evaluate three different averaging methods for processing of electroencephalogram (EEG) event related potentials (ERPs) measured from scalp in response to repeated stimulus. In ERP applications, arithmetic mean (AM) is normally employed in processing the ERPs prior to ERP detection, whereas also other averaging methods might have beneficial properties. Fast ERP detection is essential, for example, in brain computer interfaces and during spine surgery. Thus, it is of interest to search for methods to aid in detecting ERPs with as few stimulus repetitions as possible. Here, noise reduction properties of AM, geometric mean (GM), and harmonic mean (HM) are demonstrated with simulations, and ERP processing by the three methods is illustrated by processing real visual evoked potentials (VEPs).

Text-independent Speaker Identification Using Fisher Discrimination Dictionary Learning Method

In last decades, text-independent speaker recognition is a hot research topic attracted many researchers. In this paper, we proposed to apply the Fisher discrimination dictionary learning method to identify the text-independent speaker recognition. The feature used in classification is the Gaussian Mixture Model super vector. The proposed method is evaluated with public ally available dataset TIMIT. Experimental results show that the proposed method outperforms the Sparse Representation Classifier used for text-independent speaker recognition in both clean and noisy condition.

Software Fault Prediction Framework Based on aiNet Algorithm

Software fault prediction techniques are helpful in developing dependable software. In this paper, we proposed a novel framework that integrates testing and prediction process for unit testing prediction. Because high fault prone metrical data are much scattered and multi-centers can represent the whole dataset better, we used artificial immune network (aiNet) algorithm to extract and simplify data from the modules that have been tested, then generated multi-centers for each network by Hierarchical Clustering. The proposed framework acquires information along with the testing process timely and adjusts the network generated by aiNet algorithm dynamically. Experimental results show that higher accuracy can be obtained by using the proposed framework.