Local Feature Based Mage Retrieval For Network Community

At present,the multimedia data in the network community,such as images,audio and video increased dramatically.The management of multimedia information in the network community plays a vital role in information acquisition,data mining,and the management of the network communication environment.The multimedia data in the network community are mainly images.Image retrieval can not only quickly locate the required images from massive databases,but also help greatly to create a clean and healthy network community for the content management of the network community.The image of network community has the characteristics of large amount of data,complicated content and fast,accurate and comprehensive retrieval,which makes the retrieval of images in the network community very difficult.In view of the difficulties existing in the network community,this paper studies the fast and efficient retrieval of large-scale images in the network community,including image feature extraction,feature aggregation and fast and accurate retrieval.In order to solve the problem of the diversity of image public opinion in the network community and the complicated image content,this paper analyzes and compares the image feature extraction methods and selects the local feature,SIFT feature as the content feature of the image.extracting the SIFT feature of the image,and using the vector to represent the image content.In order to aggregate the local features of the image into a unified dimension vector and facilitate the indexing and searching,the Vectors of Locally Aggregated Descriptors(VLAD)are selected to encode the original local features.In order to achieve a fast retrieval of large-scale images,a good balance between storage space and retrieval speed is achieved.Using Product Quantization(PQ)to construct the index of image feature data and Asymmetric Distance Computation(ADC)to retrieval.In order to improve the recall of PQ and not reduce the retrieval speed,analyze the factors that affect the recall and retrieval speed in PQ.And improve the product quantization,a soft assignment for boundary points algorithm is proposed.The algorithm calculates the vectors near the boundary and assigns the vectors softly.Which avoids the unnecessary soft assignments in the traditional methods,so as to ensure a certain retrieval speed,and greatly improves recall.Finally,the network community image retrieval system is used to verify the effectiveness of the algorithm,which verifies the validity of the algorithm for soft assignment of boundary points.It improves the retrieval recall with a certain retrieval speed and has a good application effect.