Design And Implementation Of Cloud-based Rapid Object Recognition

The topic comes from the Research of Visual Search Capability in Cloud Compute (iSimilar). ISimilar is visual search based on the cloud computing. It is based on the Map Reduce for distributed computing, solving the large scale extended problems; It can resolve the classification, identification, nearest neighbor search using the machine learning; at the same time the outline, texture, color, key image features as to be the input of the machine learning.The main work of this paper establishes a retrieval prototype system in which the mobile is seen as the terminal, Mode servers are seen as cloud. The users can shoot the film poster or book picture then upload to the server. This system extract the features of the picture, then do the similarity search basis on this features to get the comment information of the similar pictures. The information returns to the user in form of the JSP page. The main work includes the following four aspects. The first aspect, this paper aims to build a system that solves the books and posters exact match problem. Though analysis and experiments. This paper find an exact match based on the algorithm called Surf. The second aspect, Surf methods have good results when querying image. But it takes a long time, so it cited the LSH technique to speed up the query process. The third aspect, after find the similar feature point, this paper adds the color information around it. This method improves the search accuracy to some extent. The forth aspect, this paper use the Locality-constrained Linear Coding (LLC) to solve the problem, the Visual Word ignore the relationship between the object in the picture.This paper constructs a system based on the development of software engineering standards and processes to design the system through the object-oriented thinking. The system uses the JNI technology to facilitate the extraction of the image features. Using Ajax, the user's browser experience more smooth.