An image automatic annotation algorithm

The present invention discloses an image automatic annotation method comprising the steps of: step (1) the image feature extraction data set: the underlying information obtained by image data of image feature extraction set; Step (2) Select the image training set: by select the most authoritative standard data sets of the image automatic annotation training algorithm, the dataset to contain a variety of image features and the most abundant resources, select a data set from the n data as training samples, wherein the portion of the sample is labeled, the remaining sample no labeling; step (3) the image annotation algorithms training: the sample was subjected to feature selection, and using constraint term to optimize the annotation result; step (4) the image automatic annotation:, prediction label to process by selecting the threshold value. Image tagging method based on structural features of the present invention is selected sparse realizes the image automatic annotation, groundbreaking.