K-Nearest Neighbors Relevance Annotation Model for Distance Education

With the rapid development of Internet technologies, distance education has become a popular educational mode. In this paper, the authors propose an online image automatic annotation distance education system, which could effectively help children learn interrelations between image content and corresponding keywords. Image automatic annotation is a significant problem in image retrieval and image understanding. The authors propose a K-Nearest Neighbors Relevance model, which combines KNN method with relevance models. The model solves the problems of high computational complexity and annotation results affected by irrelevant training images when joint generation probabilities between visual areas and keywords are calculated. The authors also propose a multi-scale windows method and nearest-neighbors weighting method based on rank-weighting and distance-weighting. Experiments conducted on Corel datasets verify that the K-Nearest Neighbors Relevance model is quite effective.