A Cross-Modal System for Cell Migration Image Annotation and Retrieval

A multinet system, comprising SOM's linked via Hebbian connections, has been designed and implemented for automatically annotating and retrieving cell migration images. The collateral compound keywords used in image captions and elsewhere in the text were used to train one SOM and colour moments of the image were used to train another SOM. A corpus of 1004 complex images and collateral texts was collated and the multinet system was trained to learn the association between the keywords and visual features. A test on a trial run of a smaller set of images (73 in all) showed that the multinet system outperformed single net systems and the k-means clustering.