历史信息学习的相关反馈图像检索

Image Retrieval Using Relevance Feedback with Historical Information Learning

相关反馈技术被有效的应用于基于内容的图像检索.传统的相关反馈未能充分利用检索的历史信息.为了进一步提高检索的效率与准确性,提出一种基于历史检索信息学习的相关反馈检索方法.该方法将每次检索的结果作为历史检索信息保存.进行新的检索时,判断当前查询图像与历史检索信息的语义相关性,预测检索结果,以期减少相关反馈次数.对包含8000幅图像的图像库实验表明,与传统相关反馈技术相比,该方法明显的改善了检索性能.

The relevance feedback techniques are applied efficiently to the content-based image similarity retrieval （CBIR）. How- ever ,the traditional relevance feedback techniques do not sufficiently utilize the historical retrieval information. In order to en- hance the image retrieval efficiency and precision further, a retrieval approach using the relevance feedback based on the historical retrieval information learning was presented in this paper. Every image retrieval result was saved as the historical retrieval information. When carrying out a new retrieval,decide the semantic similarity between the current query sample and the historical retrieval information,and predict the retrieval results, so as to reduce the interactive number of the relevance feedbacks. The performance of the approach was tested using an image database containing 8000 images. Results show the improved performance compared with the CBIR system with the traditional relevance feedback technique when using the same image similarity measure.