CAJ | 학술논문

针对半调图像分类问题,提出黎曼流形上的协方差建模方法和贝叶斯分类策略。根据半调图像傅立叶频谱的特点,提出一种基于模板矩阵的特征获取方法,并结合频谱信息形成协方差矩阵描述方法。通过引入有效图像判决规则和分块技术,提出一种协方差矩阵提取算法。利用样本的局部特性和核密度估计方法,实现黎曼流形上的贝叶斯分类策略。实验中研究阈值参数的选择策略,与5个相似方法进行分类性能比较,探讨有关参数对性能的影响。实验结果表明,所提出的方法在Q=32或64和L=10~15时其分类错误率低于4%,建模时间开销低于100ms,且优于5个相似方法。
침대반조도상분류문제,제출려만류형상적협방차건모방법화패협사분류책략。근거반조도상부립협빈보적특점,제출일충기우모판구진적특정획취방법,병결합빈보신식형성협방차구진묘술방법。통과인입유효도상판결규칙화분괴기술,제출일충협방차구진제취산법。이용양본적국부특성화핵밀도고계방법,실현려만류형상적패협사분류책략。실험중연구역치삼수적선택책략,여5개상사방법진행분류성능비교,탐토유관삼수대성능적영향。실험결과표명,소제출적방법재Q=32혹64화L=10~15시기분류착오솔저우4%,건모시간개소저우100ms,차우우5개상사방법。
A covariance modeling method and a Bayesian method on Riemannian manifold are presented for classification of halftone image. According to the Fourier spectrum characteristic of halftone image, a feature extraction based on template matrices is presented to form a covariance matrix by combining with the spectrum of halftone image. An algorithm for covariance matrix extraction of halftone image is proposed by introducing a decision rule of effective image and partitioning technology. A Bayesian rule based on neighbor characteristic of tested samples and kernel density estimation is presented on Riemannian manifolds of symmetric positive definite matrices. In experiments, the problem of selection on threshold parameter is studied by statistical methods, the comparisons of the proposed method with 5 similar methods are conducted, and the influences of two parameters on classification performance and time cost of feature modeling are discussed. The experimental results show that the classification error of the proposed method is below 4% and computation time of modeling is under 100 ms if parameters Q=32 or 64 and L=10~15. Furthermore, the proposed method is superior to other 5 methods.