Shaomin Zhang

Medical Image Retrieval Using SIFT Feature

In this paper, authors study using SIFT feature for medical image retrieval. Under bag-of-words framework, this paper study different methods forming visual words from local SIFT feature, and show good choice for medical image retrieval, in different condition. Through experimentation on different body parts, rotation and scale, the methods show different performance, thus provide reasonable choice for medical image retrieval. Keywordscontent based image retrieval; SIFT; visual words

Phase Space Reconstruction of Nonlinear Time Series Based on Kernel Method

A phase space reconstruction method KPCA-CA was proposed based on kernel principal component analysis (KPCA) and correlation analysis (CA) for nonlinear time series. On the basis of KPCA, the correlation was analyzed between every kernel principal component and output variable, and some kernel principal components were discontinuously chosen according to their correlation degree to form the phase space of nonlinear time series. The method was compared with other methods of phase space reconstruction. The experimental results show that modeling accuracy for nonlinear time series is highest based on the phase space reconstruction method proposed by the paper, proving the efficiency of the method

Moving Object Detection Using an In-Vehicle Fish-Eye Camera

This paper proposes a robust rear-view camera based moving object detection algorithm for backup aid and parking assist applications. A single fish-eye camera is used in order to get much larger FOV (field of view) for object detection. To detect various moving objects such as vehicles and pedestrians, the ego-motion of host vehicle is firstly estimated by A robust NGPP (near ground point projection) method. Then a novel point based moving object detection method is proposed which can detect fast motion as well as slight motion in the fish-eye image. Finally, a region based motion compensation method is used in order to filter out the false detection results caused by the error matching points. Experimental results under various conditions show that most of moving objects can be detected correctly by our algorithm.

Minimum Spanning Tree Hierarchically Fusing Multi-feature Points and High-Dimensional Features for Medical Image Registration

In this paper, we propose a novel medical registration approach based on minimal spanning tree. The proposed approach has the following contributions. (1) Compared with single type of feature points, we extracted corner-like and edge-like points from image, and added a few random points to cover the low contrast regions. (2) Instead of fixing the multi-feature points in the whole procedure, they are hierarchically updated at different registration stages. (3) Based on the feature points, in addition to using pixel intensity, we also added region based feature to include more spatial information. The proposed method is evaluated by performing registration experiments on Brain Web. The experimental results show that the proposed method achieves better robustness while maintaining good registration accuracy, compared to the conventional normalized mutual information (NMI) based registration method.

A New Two-Step Method for Medical Image Retrieval

In this paper, authors propose a new two-step medical image retrieval method which combines co-occurrence matrix texture feature and gradient phase mutual information(GP_MI). The new method provides fast retrieval respond speed while achieves high precision and good robustness in retrieval process. Experimental results show that the algorithm has potential practical values for clinical routine application.

Learning Based Combining Different Features for Medical Image Retrieval

In this paper, authors propose a new learning based method for medical image retrieval which is based on fusing different features by linearly combining different similarities. Considering the abundant classes of medical images, this paper avoid to train a classifier for each class by using large amount training data. Instead, by using optimization method to combine different features’ similarity, new method can get good performance while has no much training computation. Experimental results show that the algorithm has potential practical values for clinical routine application.

A Parameter Choosing Method of SVR for Time Series Prediction

It is important to choose good parameters in support vector regression (SVR) modeling. Choosing different parameters will influence the accuracy of SVR models. This paper proposes a parameter choosing method of SVR models for time series prediction. In the light of data features of time series, the paper improves the traditional cross-validation method, and combines the improved cross-validation with epsilon-weighed SVR in order to get good parameters of models. The experiments show that the method is effective for time series prediction.

A study of modelling non-stationary time series using support vector machines with fuzzy segmentation information

We present a new approach for modelling non-stationary time series, which combines multi-SVR and fuzzy segmentation. Following the idea of Janos Abonyi [11] where an algorithm of fuzzy segmentation was applied to time series, in this article we modify it and unite the segmentation and multi-SVR with a heuristic weighting on e. Experimental results showing its practical viability are presented.