In recent years, semantic content analysis for the construction of video information systems has been a major research topic in the diverse multimedia communities. In this paper, a novel interactive object recognition approach for Real-Time-Type (RTT) video transmission is proposed to generate semantic content with the video character as the basic unit based on low-level visual features. The basic idea is the model matching between character models and the dress features of the video character, which is specified by user interaction. A stochastic skin-color model is used for skin color classification and a knowledge-based approach is proposed to estimate the dress region of the specified character. Furthermore, a novel practical application, called Video Characters' Popularity Voting System (VCPVS), is constructed by applying the RTT object recognition approach and implemented to show the practicality of the proposed approach.