Papers by Keyword: Feature Fusion

Paper TitlePage

Authors: Qiang Chen, Yong Mei Yu
Abstract: Existing 3D model retrieval techniques focus on the global feature more than local detailed feature are not applicable to 3D CAD models. This paper combines the local feature and global feature to satisfy the needs of mechanical design. We first analyze the limitations of Shape Description (SD) algorithm. Then propose an automatic feature extraction technology based on the local characteristics of the curvature distribution. Next we fuse the extracted feature using a novel method. Lastly, we design an improved K-nearest neighbor algorithm to retrieve models. Experimental results indicate the efficiency and feasibility of the proposed method.
Authors: Wan Qing Wang, Deng Yin Zhang, Guang Shuai Shi
Abstract: Due to lack of generalanalysis method in video digital steganalysis research area, a blind detection method which is based onfeature fusion aimed at the video steganography is proposed in this paper.Compared with special steganalysis method, the method has better detection rate,lower false alarm rate, and more extensive applicability.
Authors: Xiao Ling Yao, Yan Ni Wang
Abstract: Based on the simple transfer function design, this thesis presents a technology for complicated transfer function design. It converts complicated transfer function design problem into the fusing of several simple transfer function. The keystone is to formulate the transfer function fusing problem into searching for optimal fusing proportion, and to generate the new fusing proportion using a similarity evaluation method, which is based on expectation fitness. To a large extent, it simplifies the design process of complicated transfer function.
Authors: Xiao Ke Zhu, Xiao Pan Chen, Fan Zhang
Abstract: In order to enhance the accuracy of gait recognition, a new gait feature extraction algorithm is proposed. Firstly, the gait images are preprocessed to extract moving objects, including background modeling, moving object extracting and morphological processing. Secondly, an equidistant slicing curve model based on system of polar coordinate is designed to slice the moving object, and the slicing vector is used to describe the spatial feature; Thirdly, the slicing vector is converted into frequency signal by Fourier transform to extract the frequency feature. Finally, the above two features are fused and used for the classification. The experimental results show that proposed algorithm provides higher correct classification rate than the algorithms using single feature, and meets the requirements of the real-time.
Authors: Xiao Ke Zhu, Fan Zhang, Rui Li, Yan Bin Cui
Abstract: In order to improve the classification rate of gait recognition, a new gait recognition algorithm is proposed. Firstly, the gait images are preprocessed, and the outlines of gait images are extracted and normalized. Secondly, wavelet moments of the outlines are calculated to describe the static feature of the gait images. Thirdly, the leg double triangle model is built. The first triangle consists of the mid-point of the two hips, left knee point and right knee point, and the other one consists of the mid-point of the two hips, left ankle point and right ankle point. Then the parameters of two triangles are extracted to describe the dynamic features of the gait images. Finally, the above two features are fused and used for the classification. The experimental results show that proposed algorithm provides higher correct classification rate than the algorithms using single feature, and meets the requirements of the real-time.
Authors: Ming Jie Zhang, Bao Sheng Kang
Abstract: In order to improve the robustness of visual tracking in complex environments, a novel multi-feature fusion tracking method based on mean shift and particle filter is proposed. In the proposed method, the color and shape information are adaptively fused to represent the target observation, and incorporating mean shift method into particle filter method. The method can overcome the degeneracy problem of particle. Experimental results demonstrate that this method can improve stability and accuracy of tracking.
Authors: Zhen Yu He
Abstract: In this paper, a new feature fusion method for Handwritten Character Recognition based on single tri-axis accelerometer has been proposed. The process can be explained as follows: firstly, the short-time energy (STE) features are extracted from accelerometer data. Secondly, the Frequency-domain feature namely Fast Fourier transform Coefficient (FFT) are also extracted. Finally, these two categories features are fused together and the principal component analysis (PCA) is employed to reduce the dimension of the fusion feature. Recognition of the gestures is performed with Multi-class Support Vector Machine. The average recognition results of ten Arabic numerals using the proposed fusion feature are 84.6%, which are better than only using STE or FFT feature. The performance of experimental results show that gesture-based interaction can be used as a novel human computer interaction for consumer electronics and mobile device.
Authors: Ying Xia, Xin Hao Xu
Abstract: Accuracy and stability is crucial for dynamic object tracking. Considering the scale invariance, rotational invariance and strong anti-jamming capability of KAZE features, a method of dynamic object tracking based on KAZE features and particle filter is proposed. This method obtains the global color features of the dynamic object appearance and extracts its local KAZE features to construct the object model first, and then performs dynamic tracking by particle filter. Experimental results demonstrate the accuracy and stability of the proposed method.
Authors: Jian Li, Hai Fen Chen, Li Juan Wang, Cheng Yan Zhang
Abstract: In this paper, the Gabor fused features are combined with multi-level histogram sequence to extract facial features in order to overcome the disadvantage of traditional Gabor filter bank, whose high-dimensional Gabor features are redundant and the global features representation capacity is poor. First, we get the standard face by face detection, eyes location, geometric normalization and illumination normalization. Second, to extract the multi-orientation information and reduce the dimension of the features, a fusion rule is proposed to fuse the original Gabor features of the same scale into a single feature, and then the fused image will be divided into multi-level changeable units, and the histogram of each unit is computed and combined as facial features. Experimental results on ORL via MATLAB show an encouraging performance for face recognition.
Authors: Lei Gang, Zhang Yong
Abstract: Canonical Correlation Analysis (CCA) is a powerful multi-mode feature fusion method, but in traditional CCA, the optimization function is to find a pair of projections which make the mappings of the observations of the same pattern have the maximum correlation. It is an unsupervised subspace learning algorithm. This paper, we propose a Improved Canonical Correlation Analysis (ICCA) method which improves the optimization function by adopting the supervised relationships of the patterns belong to the same class. Our proposed algorithm is validated by the experiments on Jaffe facial expression database. Be compare with other methods, the recognition rate of our method is far higher than the PCA algorithm which only adopts single-mode image features and the traditional multi-mode CCA feature fusion algorithm.
Showing 1 to 10 of 20 Paper Titles