Key Technology Research On Ancient Coins Images Using Content-Based Image Retrieval

By summarizing the relevant literature and systematically analyzing the existing key technology, this paper established a color image multi-scale edge detection model in which we combined wavelet transform with color space of color-images. Based on the model, we constructed an multi-scale feature-based image recognition and retrieval system of ancient coins image, described in detail the methods including the image noise reduction and detail enhancement, the recognition and classification of numismatic similar typecasts, multi-scale relative moments and the use of multi-scale local features ancient of coins identification and retrieval, etc.The thesis research contributions are as follows:1. As establishing an ancient coin image database, which covers the image information of the very representative of unearthed ancient coins ranged from Qin Dynasty to the Qing Dynasty, and realizing all-aspect-digital-information storage and management of the ancient coin images, we defined the database recursively to simplify the design of database table and reduce the amount of data storage.2. The article proposed a color image multi-scale edge detection model with a combination of wavelet transform and color image color space. Aiming at color image multi-scale smooth, we took wavelet transform to conduct the component filtering polished output of color images, and perform extended vector image gradient with polished images to acquire the description and expression of edge features in the image at different scales. The algorithm can extract the Numismatic color image from fine to coarse contour edge information at different scales, easy-to-the Numismatic image classification and retrieval.3. To reduce noise of color image and preserve edge detail simultaneously, we propose a color image edge multi-scale detection algorithm with enhanced detail and noise reduction to select the edge threshold of acquired edge-images based on improved soft threshold filtering function. While reducing noise, it also enhanced the edge of the reservation details realized different scale edge image fusion. The algorithm can effectively suppress noise and enhance the image details.4. Put forward an ancient-coin-image multi-scale relative moments retrieval method, which extracted relative moments feature at different scales, normalized feature vector and performed similarity measure. The method has obtained the good retrieval effect and has good versatility and robustness.5. Typecasting is one of the most important features considered in the treatment of ancient coin recognition process.To solve the problem of similar typecasting recognition, we proposed a multi-scale recognition method combined PCA with SVM. Using the extracted multi-scale edge characteristics to form the combining a number of scaling the principal component with SVM and comparison of similar typecasting of coins at different scales, the method could achieve classification of similar typecasting of ancient coins and get the best scale to identify similar words.6. As the varying degrees of wear, rust of unearthed ancient coins and differences of the lighting, scale of acquired images, we proposed a multiscale image recognition method which combined KPCA with SIFT. Employing the invariant character of local features to illustrate the KPCA-SIFT descriptors and multiscale KPCA-SIFT descriptors of extracted image, and integrating local feature points at different scales to reduce dimension through the nuclear principal component analysis methods and to identify ancient coin images which was added noise, rotation, scaling scale through multi-resolution histogram algorithm. Experiment result shows the algorithm has achieved good recognition results.