SPORTS IMAGE CLASSIFICATION BASED ON HIERARCHICHAL MULTI-SVM BAGGING NETWORK

RONGRONG JI,HONGXUN YAO,PENGFEI XU,ZHEN ZHANG
DOI: https://doi.org/10.1142/9789812709677_0136
2007-01-01
Abstract:While with promising usage in many ad-hoc research fields (e.g. automatic news categorization, sports video understanding), the issue of sports image classification is difficult to tackle due to the semantic gap in image understanding. This paper proposes a hierarchical multi-SVM bagging network to address this issue. Our scheme firstly trains both intra-class and inter-class support vector machine (SVM) to induce the confidence levels of sports images by sample bootstrapping. Then an aggregation procedure is conducted on inter-class SVM by weighted majority voting, where the output of its corresponding intra-class SVM is the weight to rectify its contribution in aggregation. These voting results are utilized to classify sports images into their corresponding category. Our hierarchical multi-SVM bagging network combines both one-to-one and one-to-all SVM classifiers. It provides a good robustness to training data …
What problem does this paper attempt to address?