Research On Blue And White Porcelain Decorations Image Classification Based On Content

With the advent of the Internet, the digital image showed explosive growth, image retrieval research also will be valued. The essence and core retrieval technology is to classify a large number of images, and the traditional text-based classification method costs high labor, a large man-made factors. With the development of machine learning, content-based image classification that extract color, texture, shape and other low-level information, as a basis for conduct of classified research in recent years has become a hot spot, it can better overcome traditional classification methods defects. At the same time as the rise of digital museum for the digital management of traditional painting and calligraphy works have certain results, but most Chinese blue and white porcelain features works of art did not get special attention, compared to traditional painting and calligraphy, porcelain retrieval and classification of images is very weak.This will be blue and white decoration as a starting point, research content-based image classification method. The main difficulty is that compared to conventional natural image, how will semantic and low-level features blue and white images linked against the blue and white image of how to construct an appropriate classification and similarity measure, to reach the correct classification rate can be acceptable, and provide a suitable matching image.To solve the above problems, researched include the following:(1) In order to effectively utilize color feature an image. Study the traditional color histogram extraction algorithm defects, with blue and white porcelain decoration features, we put forward a multi-threshold quantify HSV(Hue Saturation Value)color histogram algorithm. Blue and white color values for the single, but the distribution is different color characteristics of different categories of laws, study color autocorrelation plot extraction algorithm and color feature, and the block in the form of improvements have been made. Global and eventually combine color space representation to obtain a valid set of mixed color characteristic value.(2)For the blue and white porcelain decoration in texture obvious texture features as the underlying feature an image, and streamline input final classifier. GLCM were studied, SGF(Statistical Geometrical Features), and combined with PCA dimensionality reduction optimized for Gabor wavelet transform. The characteristic values obtained are compared and screening to finally obtain a set of optimal texture feature value.(3) In order to solve the conventional image matching long distance computing time, the same amplitude and so on, select cityblock optimize obtain the most appropriate measure of distance. And use this classification to achieve the above-mentioned color and texture selection process.(4) Because SVM(Support Vector Machine) classifier advantage of small sample classification, research SVM classifier and optimizing kernel function, classification introduced to assess the effect of the confusion matrix, comparison with other optimization algorithms and analysis, the method of this paper is more effective well, more feasible. The final design and completion of blue and white porcelain decoration classification system based on a set of low-level features.