A KNN-based Chinese generating method on static image description

In this research we present a novel cognition language model ViMac. Through the vision system ViMac can acquire the semantic of the objects shown in 2D images and output its own description for the new images. ViMac uses prototypes as its semantic representation and use KNN method to measure the similarity between these data points in visual subspaces. The BLEU evaluation shows that this method can generate good performance on new image description.