Research Of Image Classification And Matching Based On Hadoop

With the rapid development of Internet technology, more and more information is obtained from the Internet to the people. These people are interested in information which is mainly pictures, video and other multimedia information. When people retrieve relevant image content, classification of image data is playing an increasingly important role. How to identify an.image category from the large number of images has become an issue in the field of image processing.Due to the complex information structure of image, when computer dealing with image requires a lot of computing resources. The rise of the concept of cloud computing makes distributed storage and computing receive extensive attention. The open source Hadoop distributed framework came into being in this environment. Hadoop is mainly composed of two parts, namely the Hadoop Distributed File System (HDFS) and MapReduce parallel programming framework. The emergence of Hadoop greatly simplify distributed programming, HDFS used for storage and retrieval of massive data, MapReduce used for data parallel computing. Users can just focus on writing Map and Renduce functions to realize their service logic, Regardless of cluster data communication and task allocation problem between the various nodes. In this paper, Hadoop distributed computing framework is used to implement image processing.Details are as follows.Firstly, in this article, HDFS and MapReduce which are based on Hadoop distributed platform have been studied deeply. Meanwhile, the image features extraction and matching algorithm in image classification is also the focus of our study. Secondly, we have built a Hadoop distributed platform on Linux operating system, and base on this platform design and implement an image classification system. The system is based on Hadoop platform, we just focus on writing Map and Renduce functions to realize their service logic, Regardless of cluster data communication and task allocation problem between the various nodes. so it has good fault tolerance and scalability.Finally, test results show that when the system dealing with a large number of image data, whether the ability of images storage or the speed of image classification has significantly improved compared to the traditional single node image classification system.