An SVM-based soccer video shot classification

Video shot classification is an important component of content-based video retrieval. It is also a basic step towards video abstract, event detection and content filtering. According to the characters of shots for soccer videos, an integrating color and edge distribution shot classification method is presented. First, A GMM model is trained for grass pixel. And then the grass distribution is computed based on the trained GMM. Due to the sensitivity to light, field and time for color features, edge distribution is computed through canny operator. These features are reasonable complementarities to color features and have no sensitivity to light, field and time. Integrating two types of features embodies main characters of different shot types. One-against-others SVMs are designed for shots classification. Experiments show that our method performs better results.