Research And Implementation Of Image Tag Completion Based On Sparse Representation Theory

With the rapid development of the internet,the number of online community images are exploding quickly,how to manage the image dataset and retrieval image become a big problems.Tagging is becoming an important method to annotate the images.However,because of the difference of the network user's cultural background,different user may add different tags on the same image,some correct tags about image may be missed,hence it is difficult for user to search a image form large image dataset,in order to increase the number of correct tags on images,tag completion become a hot topic in image retrieval.So we proposed the image tag completion based on sparse representation theory.By researching and analyzing about existed method,we found that the existed methods about tag completion,they usually perform global refinement for the tagging matrix,we propose a new method,in the methed we use each row vector rebuild image specific,use group sparsity as weight for feature selection.Then,we use image-tag matrix as tag information,merge result of reconstruction with initial matrix,consider the initial tag information,we can add correct tags on images.Before reconstruction,we can use KNN method to find K most similar image,use these image build the dictionary matrix for image-specific reconstruction.For the final result,we choose the tag which has high value of reconstruction label on image.The proposed method use group sparsity in tag completion framework,provide a new method of tag completion,it has important significance,in order to estimate performance of proposed algorithm,we use Corel5k and Flicker30 to estimate the result of experment.