Abstract: In the training process of machine vision system, a large number of labeled images are needed to enhance the recognition ability. The traditional method is to gather a group of people for independent annotation, which is inefficient and of poor quality. This paper designs an image labeling system based on crowdsourcing. The system uses collaborative filtering technology to push pictures to volunteers with corresponding majors or interests, and then sorts and classifies the tag set of the same picture through semantic processing algorithm, and finally obtains accurate and effective tags. Experimental results show that the system has better robustness and higher efficiency than traditional image labeling methods.

Key words: image labeling system, collaborative filtering algorithm, tag semantic processing, tag recommendation algorithm