全局及其个性化区域特征的图像检索

交通领域个性化图像检索的关键是根据业务需求通过重点监控车辆的个性化特征 在海量数据库中进行匹配,其目的是捕获与重点监控车辆相关的卡口信息。目前的图像检索算法包括基于文本的图像检索、基于内容的图像检索方法和基于语义的图像检索。针对交通领域的图像检索需求,提出了一种基于全局以及个性化感兴趣区域特征的图像检索算法。通过使用交通图像库进行检索验证,对个性化特征进行精准滤除,从而得到准确的检索结果。实验表明,此种基于全局特征结合个性化感兴趣区域特征的图像检索算法解决了CNN高层特征对个性化局部特征描述能力低、检索耗时等问题,并通过个性化局部特征提高了检索效果,使得检索率、平均准确率都达到90%,呈现出较好的检索效果,计算速度快,具有较强的鲁棒性和实用性。

Image Retrieval of Global and Personalized ROI Adjustment of Features

The key of personalized image retrieval in the area of traffic is to capture the bayonet related information of the key monitoring vehicle through their personalized features.The current image retrieval algorithms include text based image retrieval,content-based image retrieval and semantic based image retrieval.Based on the requirements of traffic in the field of image retrieval,a new image retrieval algorithm based on global and individual interest region features was presented.To get accurate search results,we used the traffic image database to search,verify and filter the accurate personalized features.The experiment demonstrates that this method solves the problem of CNN features,such as low abilityto describe personalized features and time-consuming.This method also has strong ability to describe personalized features,both the image retrieval rate and the average accuracy rate reach 90%,showing a better retrieval effect,fast calculation speed,strong robustness and practicality.