摘要
研究一种电视图像目标实时分割和识别算法.在二维图像不变矩和相对矩的基础上,进一步组合优化得出4个不变矩,结合复数矩、圆方差和椭圆方差组成目标特征向量,利用k-近邻法实现目标的识别和分类.图像分割采用改进矩不变阈值分割和基于梯度的自适应阈值分割提取目标.仿真实验表明,提取的目标特征量对于平移、缩放和旋转均能保持较好的不变性.用该分割算法分割的图像边缘清晰,分割时间为8 m s,易于硬件实现.
The real-time image segmentation and recognition algorithm for a TV seeker is developed. Based on Hu＇s invariant moments and relative moments, a novel pattern recognition method is presented, which can classify the targets by using the four invariant moments after further optimized and combine, together with the complex moment, circularity variance and ellipse variance to construct the eigenvector of the image, the improved moment-preserving thresholding method and the adaptive gradient thresholding method are used to pick-up targets. The results of experiments show that the target eigenvectors have the property of translation, rotation and scaling invariance. The segmentation algorithm can segment the images automatically, completely and rapidly its hardware is easy.
出处
《北京理工大学学报》
EI
CAS
CSCD
北大核心
2005年第9期786-790,共5页
Transactions of Beijing Institute of Technology
基金
国家部委预研项目(51405030104BQ0171)
关键词
图像分割
特征提取
目标识别
组合不变矩
image segmentation
feature extraction
target recognition
combined invariantmoment
