出版社：The Institute of Image Information and Television Engineers
摘要：A method is proposed for automatically generating an image-retrieval algorithm corresponding to keywords in a quary. It uses vector quantizers and an N-out-of-M detector in the feature space. The objects corresponding to a keyword are assumed to have common features, and the background image is considered to be random noise. The parameters for the vecotor quantizers and detector are determined through training. First, all quantizers are assumed to be scalars, and the parameters are determined using the genetic algorithm. These quantizers are then grouped into vectors quantizers, and their parameters are optimized by clustering. A practical experiment showed that the proposesd method can retrieve images corresponding to keywords.
