Biography
Biography: HaoWu
Abstract
Supervised retrieval model has been widely used in the field of computer vision and its high-quality result is supported by enough learning instances .However, in the process of experiments, it’s difficult to offer enough learning instances for each category. Especially for some special categories, the drawback is more obvious. So how to solve the problem has become one challenging problem.
For this problem, we proposed one new model that can use candidate learning instances to replace the learning instances (In this paper, we mainly consider positive instances). On the one hand, the improved spatial pyramid matching function contributes to retrieve candidate learning instances effectively. On the other hand, an optimized SVM model make the most of candidate learning instances to keep the accuracy of retrieval. At last, we did enough groups of experiments using the new model. The experimental results show that our new model not only can reduce the number of learning instances but also can keep the high-quality of retrieval.
