纸基纳米金阵列图像特征值自动提取方法

Automatic Extraction Method for Feature of Nanogold-Based Membranes Array Image

纸基纳米金作为一种能够实现重金属离子检测的比色传感器,具备快速、精确、特异性高等优点。当前,手动处理是实现纸基特征信息提取的主要方式,其效率低下,容易引入人为误差,所得特征信息的可重复性也较差。针对这些问题,在纸基阵列图像基础上,提出基于HSI颜色空间和种子区域生长的两阶段自动特征提取方法。首先将图像由RGB空间转换到HSI空间完成粗分割、形态学滤波和网格划分;在此基础上提出首先基于色调吸引力(HAF)和亮度吸引力(IAF)的相似性度量的种子区域生长(SRG)算法,该算法有效地结合SRG算法的像素物理位置邻接特性和HAF及IAF的颜色空间邻接特性,实现细分割和特征提取。实验采用10张5×5纸基阵列(共计250个反应点),对该特征提取方法进行精度和稳定性测试,并采用10张不同尺寸的纸基阵列,对其进行自适应特性测试。实验结果表明,该特征提取方法的平均误差在1%以内,最大误差及标准差在5%以内,针对不同尺寸的阵列处理正确率为100%,具有较高的精度、稳定性和自适应性。

As a colorimetric sensor that is able to detect heavy metal ion,nanogold-based membranes have the advantages of high speed,accuracy and high specificity.Currently,processing image one by one manually is the main method used for extracting the sensor′s color features,which is inefficient,is easy to introduce human error,and the reproducibility of the acquired feature information is also poor.In order to solve these problems,here we proposed a two-stage automatic feature extraction method based on HSI color space and seeded region growing(SRG)on the basis of array image of the nanogold-based membrane.At the first stage,the array image was transformed from RGB space to HSI space to complete rough segmentation,spots-gridding and filtering;based on the first stage,the second stage proposed a seeded region growing(SRG)algorithm with similarity measurement based on hue attractive force(HAF)and intensity attractive force(IAF),which combined the adjacency characteristic of physical location of pixel of SRG algorithm and adjacency characteristic of color space of HAF and IAF to realize the precise segmentation and feature extraction effectively.In the experiment,10 sheets of 5*5 paper-based arrays(250 reaction points in total)were used to test the accuracy and stability of this feature extraction method,and 10 sheets of paper-based arrays of different sizes were used to test the self-adaptation characteristics.The results of the experiment showed that the average error of the feature extraction method was less than 1%,maximum error and standard error was less than 5%,and the correct rate was 100%for arrays of different sizes.In conclusion,the proposed method was high in accuracy,stability,and adaptability.