基于混合特征的互联网茄子图像检索方法与系统

发布时间：2018-06-10 11:02

本文选题：图像检索 + 特征提取　；参考：《农业工程学报》2017年S1期

【摘要】：互联网图像数据的爆炸式增长使得有效检索变得越来越重要,不同于文字的检索,有效的图像检索仍然是一个开放的问题。该文提出了基于混合特征的互联网茄子类图像的检索方法,并开发了图像检索系统。该文采用Hu不变矩(hu invariant distance)作为几何不变特征,采用颜色矩方法,通过计算HSV空间的三阶矩来描述颜色特征,采用分水岭算法(watershed algorithm)提取茄子的轮廓特征,通过长宽比特征区分长茄和圆茄,最后综合几何不变特征、颜色、轮廓特征进行茄子对象的描述,分别给3种特征赋不同的权重,形成检索系统。试验验证,该文方法和系统在测试数据集上查全率为87.6%,查准率为87.6%,相比于只采用Hu不变矩方法(其查全率为31.75%,查准率为31.75%)在Hu不变矩加颜色特征方法(查全率为52.8%,查准率为52.8%),有了一定的提升,验证了方法的有效性。
[Abstract]:The explosive growth of Internet image data makes effective retrieval more and more important. Different from text retrieval, effective image retrieval is still an open problem. In this paper, a hybrid feature based image retrieval method for eggplant is proposed, and an image retrieval system is developed. In this paper, Hu invariant is used as geometric invariant feature, color moment method is used to describe the color feature by calculating the third-order moments in HSVspace, and watershed algorithm is used to extract the outline feature of eggplant. The length to width ratio feature is used to distinguish the long and round eggplant. Finally, the geometric invariant feature, the color feature and the contour feature are integrated to describe the eggplant object. The three features are given different weights to form a retrieval system. Test verification, In this paper, the recall rate on the test data set is 87.6, and the precision is 87.6. Compared with the Hu invariant moment method (its recall rate is 31.75 and precision is 31.75), the Hu invariant moment plus color feature method (the recall rate is 52.8, the precision is the same as the Hu invariant moment method). 52.8%, there has been a certain promotion, The validity of the method is verified.
【作者单位】：中国农业大学信息与电气工程学院;北京市农业物联网工程技术研究中心;农业部农业信息获取技术重点实验室;鲁东大学信息与电气工程学院;
【基金】：国家自然科学基金(61472172,61100115) 国家国际科技合作专项项目(2015DFA00530)
【分类号】：TP391.41

【相似文献】