弦特征矩阵:一种有效的用于植物叶片图像分类和检索的形状描述子

发布时间：2018-05-16 21:46

本文选题：形状描述 + 内外弦长　；参考：《计算机学报》2017年11期

【摘要】：植物叶片形状一般具有小的类间差异和大的类内变化,再加之叶片的自遮挡和噪声的干扰,给叶片形状的识别带来了很大的挑战.文中提出了一种新的形状描述子——弦特征矩阵(Chord-Features Matrices,CFM),精确而又鲁棒地应用于植物叶片图像的分类和检索问题.该方法将目标轮廓线的弦,依据轮廓线所围成的区域,分成内和外两个部分.因其与轮廓线凸凹特性的相关性,该方法用多个尺度级的内部弦长和外部弦长生成两个矩阵,旨在隐式地描述轮廓线的多尺度的凸凹特性.该方法还定义了多个尺度级的弧到弦的平均投影长度,并构成矩阵,以反应轮廓线在各个尺度级下的弯曲程度.组合这3个矩阵所形成的CFM描述子,全面地刻画了轮廓线的几何特性,具有强的形状表征能力.用Swedish、Flavia和ImageCLEF这3个具有挑战性的叶片图像测试集,对文中提出的CFM方法分别进行分类和检索性能评估.实验结果表明,文中提出的方法在精确度和对噪声的鲁棒性方面均优于其他植物叶片图像分类和检索方法.而文中提出的方法在MPEG-7测试集上的实验结果则验证了其具有应用于一般的形状识别任务的潜力.
[Abstract]:Plant leaf shape generally has small differences between classes and large intra-class changes, coupled with leaf self-occlusion and noise interference, which brings great challenges to the recognition of leaf shape. In this paper, a new shape descriptor, Chord-Features Matrices, is presented, which is applied to the classification and retrieval of plant leaf images accurately and robustly. In this method, the string of the contour line is divided into two parts according to the region formed by the contour line. Because of its correlation with the convex and concave characteristics of contour lines, the method generates two matrices by using the internal chord length and the external chord length of multiple scales to implicitly describe the multi-scale convex and concave properties of the contours. The method also defines the average projection length of arc-to-chord of multiple scales and forms a matrix to reflect the bending degree of contours at each scale level. The CFM descriptors formed by the combination of these three matrices can describe the geometric properties of the contour line comprehensively and have a strong ability of shape representation. Three challenging leaf image test sets, SwedishFlavia and ImageCLEF, are used to evaluate the classification and retrieval performance of the proposed CFM method. The experimental results show that the proposed method is superior to other methods in classification and retrieval of plant leaf images in terms of accuracy and robustness to noise. The experimental results on the MPEG-7 test set show that the proposed method has the potential to be applied to general shape recognition tasks.
【作者单位】：南京财经大学信息工程学院;南京财经大学电子商务省级重点实验室;
【基金】：国家自然科学基金(61372158) 江苏省自然科学基金(BK20141487) 江苏省“333工程”高层次人才资助项目(BRA2015351) 江苏省科技计划项目(产学研合作前瞻性联合研究)(BY2016009-03) 江苏省高校优势学科建设工程资助项目(PAPD) 江苏省高校优秀科技创新团队项目(2017-15)资助~~
【分类号】：TP391.41

【相似文献】