基于SLIC与APMMD结合的图像分割算法研究  

Research on Image Segmentation Algorithm Based on Combination of SLIC and APMMD

作　　者：傅江 FU Jiang(School of Mechanical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学机械工程学院,江苏南京210094

出　　处：《机械制造与自动化》2024年第3期141-145,共5页Machine Building & Automation

基　　金：江苏省研究生科研与实践创新计划项目(SJCX21_0127)。

摘　　要：超像素是由一系列特征相似且位置相邻的像素点组成的小区域。采用超像素分割既能降低图像分割的复杂度,又能更好地保留局部信息及边缘信息。针对管壁污渍识别和零件盒零件分类的要求,将SLIC与基于近邻传播最大-最小距离算法(APMMD)结合使用来达到更好的图像分割效果。该方法根据传统SLIC算法的步骤,在颜色空间转换时通过数据类型转换来优化内存;在均匀分配初始种子点之前,增加APMMD算法来解决初始聚类中心不合理导致聚类结果局部最优的问题,可在一定范围内防止种子点落在梯度较大的轮廓边界上。通过边界召回率和欠分割错误率验证了所提算法,发现其颜色空间转换时内存减少1.5 M,聚类时准确率提高了8.1%。A superpixel is a small region consisting of a series of pixel points with similar features and adjacent positions.The use of superpixel segmentation can both reduce the complexity of image segmentation and better preserve local information and edge information.For the requirements of tube wall stain recognition and parts box part classification,SLIC is used in combination with the nearest neighbor propagation based maximum minimum distance algorithm(APMMD)to achieve better image segmentation results.The method optimizes memory by data type conversion during color space conversion according to the steps of traditional SLIC algorithm.Before uniform assignment of initial seed points,the APMMD algorithm is added to solve the problem of unreasonable initial clustering centers leading to locally optimal clustering results,which can prevent seed points from falling on contour boundaries with large gradients within a certain range.The proposed algorithm is verified by the boundary recall rate and under-segmentation error rate,and it is found that its memory is reduced by 1.5 M during color space conversion,and the accuracy rate is improved by 8.1 percentage points during clustering.

关 键 词：超像素 聚类 SLIC APMMD 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]