Multi-Use Learning Instance for Optimized Image Retrieval

Sophisticated machine learning methods significantly improve the accuracy of image retrieval, but the majority of classic machine learning methods largely depend on adequate learning instances, especially for deep learning based methods. Once the learning instances are insufficient, the retrieval accuracy will significantly decrease. Actually, excessive attention has been paid to the training model, while the effective utilization of learning instances is neglected. Aiming at this problem, we propose one multi-use learning instance for optimized image retrieval model. Through using more comprehensive cost function, some multi-use learning instances are effectively applied in image retrieval of different categories. On the one hand, multi-use learning instances could be sufficiently utilized by different weights, reducing the number of learning instances in practical use. On the other hand, the accuracy of image retrieval is generally preserved even though multi-use learning instances are employed. Furthermore, the superiority of our method has been proved by sufficient contrast experiments.