Font Size: a A A

Research On Image Edge Detection And Image Matching And Their Applications

Posted on:2004-04-02Degree:DoctorType:Dissertation
Country:ChinaCandidate:H Y WangFull Text:PDF
GTID:1118360122975011Subject:Electrical engineering
Abstract/Summary:PDF Full Text Request
Edge is the most important characteristic of image, which includes the useful information of image recognition. The image recognition is the core section of image monitor. This paper studies the technologies of image recognition, including the image edge detection and image matching and their applications based on the project of the image monitor system of the unmanned substations. In conclusion the main contents are as follows:1. In the field of edge detection, the noise reducing and edge localization are contradictory parts and are regarded as an ill-posed problem. Based on this ill-posed problem of edge detection, the edge types that exist in real images are described as mathematical models and the edge models that smoothed by Gaussian function are regarded as the research objects. The paper systematically analyzes the characteristics of the different edge types and the relations between the localization of the different edge types and the smoothing scale while using the numerical differentiation as the method to detect edges. The conclusions can be used to classify the edge types. If the edge types are classified in prior, then the smoothing scale can be selected rightly and the ill-posed problem of edge detection will be solved comparatively well.2. The characteristics of image edges can be described by information measures quantificationally. The paper forms the three information measures into measure vector as the input of CMAC- Cerebellar Model Articulation Controller neural network and proposes a method of edge detection based on information measures and CMAC. The edge contours generated by this method is very legible. The method has a high runtime performance and improves the resistance to noise. The paper presents a real-time monitoring scheme for the porcelain bottles of the main transformer by applying this method of edge detection to unmanned substation.3. The application of wavelet transform and multiresolution analysis to edge detection is analyzed. Incorporating with the invariant moments, the paper present a real-time monitoring scheme for image monitor system of the unmanned substations by dynamically detecting the cracks of the porcelain bottles. Simulation results and applications show that the method presented is valid and effective.4. The relations of the wavelet's self-characteristics, such as symmetry, convergence,regularity and wavelet cancellations and their influences on edge detection are studied theoretically. Then the wavelet selection principles in edge detection are presented. Simulation results are coincident with the theoretical analysis. The scale selection in edge detection based on wavelet transform is studied by using a general slope edge model as an example.5. The application of Hausdorff distance to the algorithm of the image matching is researched. Based on the information measures and Hausdorff distance, a new matching strategy is proposed. The experimental results demonstrate that the proposed strategy speeds up the matching process and improves the resistance to noise. In addition, this method matches the image occlusions correctly and overcomes the mismatching problems that induced by noise, spurious edge segments and outlier points.
Keywords/Search Tags:image monitor, image edge detection, image matching, CMAC neural network, wavelet transform, wavelet selection, information measure, Hausdorff distance, crack detection
PDF Full Text Request
Related items