Image Classification Retrieval And Distributed Implementation Based On Feature Semantic

With the development in the field of computer technology, especially the development of multimedia technology, image data on the network shows an exponential growth, and peoples’demand for these multimedia image data is also growing, but with the number’s increase of images that people found it is more and more difficult to get image which they need, the vast majority of internet image information can not be used by people. How storage and retrieval of images efficiently in the Internet has become a hot research field in academia and industry, and also it is a major problem faced by the people. People in academia and industry invent a variety of image classification and retrieval technology in recent years, but the actual application effect is not very satisfactory Semantic-based image retrieval can reflect people’s subjective feeling and understanding of an image, it is a more rational approach to retrieve images. Therefore, we choose this area as our researchThis paper summarizes the research results in this field, we gave a brief description of the meaning and representation of a image semantic, selected the appropriate method to extract the underlying physical characteristics of an image, and researched the normalization and similarity measurement methods for the image underlying physical characteristics, and convert the underlying physical characteristics of the image to image semantic. On the basis of image feature extraction, by studying the application effect of different machine learning algorithm, i selected the K-nearest algorithm to train the extracted features. Because the image is characterized by high-dimensional data, training for high-dimensional data is a time-consuming task, so istudied how to use distributed platform hadoop to train images semantic features, and implemented a distributed K-nearest algorithm to speed up the data training efficiency.Completed the design of a distributed image retrieval system for book illustrations that users upload, which can accurately determine the title, summary, full text link and collection link corresponding to uploaded images from user, the system has been applied to "Classic Reading" systems in Beijing University of Posts and Telecommunications Library to facilitate the users to retrieve, which enrich the user’s retrieval approach and experience.