摘要
借助于低成本深度传感器,产生了深度与颜色同步的RGB-D图像。针对RGB-D图像分割困难以及黑猩猩优化算法精度低、收敛速度慢和易陷入局部最优的问题,提出了基于改进黑猩猩优化算法(Improved Chimp Optimization Algorithm,IChOA)的RGB-D图像核模糊聚类算法。首先,对RGB-D图像进行特征提取生成6个特征子集;其次,引入Levy飞行策略和非线性惯性权重对ChOA进行改造;最后,利用IChOA对6个特征子集进行核模糊聚类,得到多个最优聚类,然后通过聚集超像素方法对多个最优聚类进行不同组合的分割,生成最终的分割结果。采用NYU depth V2室内图像数据集进行实验,与现有的一些分割方法(阈值分割,模糊子空间聚类,残差驱动的模糊C-均值,硬C-均值,模糊C-均值,核模糊聚类,基于混沌kbest引力搜索算法和随机亨利溶解度优化算法)进行比较,结果表明所提出的RGB-D分割算法优于比较的算法。
With the help of a low-cost depth sensor,an RGB-D image with depth and color synchronization is produced.Aiming at the difficulties of RGB-D image segmentation and the problems of slow convergence speed,low accuracy and a susceptibility to getting trapped in local optima of chimp optimization algorithm.A RGB-D image kernel fuzzy clustering based on improved chimp optimization algorithm(IChOA)is proposed.Firstly,six feature subsets are extracted from RGBD images.Secondly,Levy flight strategy and Nonlinear Inertia Weight are introduced to transform ChOA.Finally,IChOA is used to kernel fuzzy clustering on six feature subsets,and multiple optimal clusters are obtained.Then the aggregating superpixels method is used to segment multiple optimal clusters in different combinations,and generated the final segmentation result.The experiment is carried out by using the indoor image dataset of NYU depth V2,and compared with some existing segmentation methods:threshold segmentation,Fuzzy subspace clustering,Residual-driven Fuzzy C-Means,hard C-means,fuzzy C-means,kernel fuzzy clustering,chaotic kbest gravitational search algorithm and random Henry gas solubility optimization algorithm.The results show that the proposed RGB-D segmentation algorithm is superior to the compared algorithms.
作者
刘恒
范九伦
郭培岩
LIU Heng;FAN Jiulun;GUO PeiYan(School of Telecommunications and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710100,China)
出处
《微电子学与计算机》
2024年第9期10-21,共12页
Microelectronics & Computer
基金
国家自然科学基金(62071378,62071379)。
关键词
RGB-D图像分割
核模糊聚类
黑猩猩优化算法
聚集超像素
RGB-D image segmentation
kernel fuzzy clustering
chimp optimization algorithm
aggregating superpixels
