摘要
针对传统物体形状分类算法中图像的空间结构特征表示不够准确,以及分类器模型参数易陷入局部最优的问题,提出结合重叠金字塔与贝叶斯优化神经网络的物体形状分类方法。首先,将物体轮廓分割为不同长度的轮廓片段作为形状的基本特征,并用局部线性编码器对其编码;然后,使用提出的空间重叠金字塔模型,将图像表示为空间金字塔直方图向量;最后,使用贝叶斯优化的前馈神经网络分类器对得到的图像表达进行分类。在常用的Animal标准图像库上实验证明,本文方法可以完整记录形状的内容和结构信息,与轮廓片段包算法相比,准确度提高了1.4%。
In order to solve the problems of traditional object classification methods, such as the inaccurate expression of spatial structure features, and the classification model parameters trapped in local optimum, we propose a method that combines the overlapping pyramid method with the Bayesian optimized neural network. Firstly, we extract the contour fragments of different lenghts from the object contour as features, and encode them with the locality-constrained linear coding encoder. Then, the proposed spatial overlapping pyramid histogram is used to represent the images. Finally, the Bayesian optimized feedforward neural network classifier is used to accomplish the classification. The experimental results based on the standard Animal dataset show that the accuracy of the proposed method is improved by 1.4% as compared to the Bag of Contour Fragment method, indicating that the proposed method can accurately represent the context and structure of the shape and is effective in object classification.
作者
张善新
范强
周治平
Zhang Shanxin;Fan Qiang;Zhou Zhiping(Engineering Research Center of Internet of Things Technology Applications, Ministry of Education Jiangnan University, Wuxi, Jiangsu 214000, China)
出处
《激光与光电子学进展》
CSCD
北大核心
2018年第6期173-178,共6页
Laser & Optoelectronics Progress
基金
江南大学自主科研基金(1252050205170640)
关键词
图像处理
形状分类
贝叶斯优化
重叠金字塔
前馈神经网络
image processing
shape classification
Bayesian optimization
overlapping pyramid
feedforward neural network
