Font Size: a A A

Research On Feature-level Image Fusion Method Based On Multi-source Image

Posted on:2015-04-16Degree:MasterType:Thesis
Country:ChinaCandidate:M M JinFull Text:PDF
GTID:2298330467955093Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
In recent years, based on information fusion theory and applications multi-sourceimage fusion technology gradually developed up and matures, and one of the mostclassic application is the fusion of visible and infrared images. Visible sensor getsimages through accessing the reflection information of target scene, and infrared sensorreceives infrared radiation from the scene object by thermal detection device to getimages. In practice, although the visible sensor has many advantages, but its images arenot clear enough in the harsh environment, and infrared sensors just can overcome thisshortcoming. Therefore, image fusion of two sensors is able to complement each other,to get a more comprehensive description of the target scene. Feature-level image fusioncompared with pixel-level fusion and decision-level fusion, it not only can retain targeteffective identification information but also remove data redundancy. It has betterclassification performance, and it is a good image fusion method. However, there is lessliterature and research on feature-level fusion, and is also actively exploring, so in thiscontext this thesis made a thorough study on infrared and visible images of feature-levelfusion method. In this thesis, the specific contents are as follows:(1)Analysis of the current commonly used feature extraction target characteristics,including GLCM, Hu invariant moments, affine invariant moments, Wavelet momentsand Zernike moments, and study the basic structure and principles of feature-levelidentification of target recognition system, and describe the process of the infrared andvisible feature-level fusion.(2) Study the independent component analysis (ICA) algorithm. Elaborate the basicprinciple and independence metrics of the ICA algorithms, focusing on the fast ICAalgorithm and its application to feature fusion of infrared and visible image.Experiments show that the fusion features for ICA method has good recognition rate.(3) Study the principal component analysis (PCA) algorithm. The PCA is appliedto feature fusion, constructing the correlation matrix, and finding its eigenvalues andeigenvectors, selecting the integration features according to the accumulative contribu-tion rate. PCA method proved to be an effective feature fusion method. (4) Study the canonical correlation analysis (CCA) algorithm and its improvements.CCA algorithm in the case of a small sample of high dimensional will face singularcovariance matrix problem, to solve this problem, we choose the PCA algorithm toimprove it. Use PCA to reduce the dimension of the data in the first, and then in a lowdimensional space using CCA method for solving integration features. Throughexperiments has verified the the feasibility of this method, and get a good recognitionrate.
Keywords/Search Tags:Image Processing, Feature Level Fusion, Principal Component Analysis, Independent Component Analysis, Canonical Correlation Analysis
PDF Full Text Request
Related items