New relevance feedback method based on dynamic weight updating

In content-based image retrieval systems using relevance feedback are often considered relevant images but ignored irrelevant images. This paper proposes an approach of updating corresponding weight by combining the proportion of relevant and irrelevant images. According the proportion, using the standard deviation adjust the feature value. To avoid the state of sub-optimal, a disturbing factor has been used to get the best feature weights. Experiments show the better result.