Per-class key feature weighting method for image retrieval

摘要

To set a weight for retrieving a large image, a key value affecting an image configuration is defined according to a visual word about an image. A cost function is determined to improve a classification between the classes of the image. A weight mapping function is determined through optimization. The large image is retrieved by re-ranking a query image after a subset is made without the weight for the query image and the weight is applied according to the class. [Reference numerals] (S100) Define the standard deviation in which the size of a visual word by class is normalized as a key value; (S110) Define a cost function by using the key value; (S120) Learn a weight mapping function through cost function optimization; (S130) Re-rank by a weight vector