基于MRF层次模型的贾卡经编针织物图像分割技术  

Segmentation of jacquard warp-knitted fabric image based on hierarchical Markov random field model

作　　者：张扬[1,2] 蒋高明[1] 姚君洲[1] 童有成[3] 

机构地区：[1]江南大学经编技术教育部工程研究中心,江苏无锡214122 [2]浙江理工大学科技与艺术学院,浙江杭州311121 [3]浙江大学宁波理工学院,浙江宁波315100

出　　处：《纺织学报》2012年第12期102-106,共5页Journal of Textile Research

基　　金：国家科技支撑项目(2012BAF13B03);江苏省产学研联合创新资金项目(BY2010119)

摘　　要：基于贾卡织物特点和层次马尔科夫随机场建模,提出一种贾卡经编针织物图像的分割方法。贾卡织物图像在扫描过程中因其立体肌理会带来噪音信息,需进行高斯低通滤波预处理,在此基础上,采用haar小波进行多尺度分解以模拟人类视觉处理信息过程。在特征场建模上采用有限通用混合模型来逼近贾卡织物图像的小波系数,同时融合零均值隐状态的混合概率分布函数来描述贾卡经编针织物图像的噪音信息,在标签场建模上采用尺度内的非因果马尔科夫和尺度间因果马尔科夫来确定先验模型,在贝叶斯准则下完成图像分割。结果表明,该方法能较好地实现贾卡经编针织物图像的纹理分割。This paper proposes an approach to segment the jacquard warp-knitted fabric image based on jacquard fabric characteristics and Markov random field theory.A Gaussian low-pass filter is adopted to decrease the noise caused by discordant light reflection due to the uneven jacquard fabric surface during scanning procedure,followed by multiresolution wavelet decomposition by Harr wavelet to simulate the process of the human visual system.Next a finite general model is presented to approximate the wavelet coefficients in feature field,containing the noise with zero-mean Gaussian distribution to describe the noise in jacquard warp-knitted fabric image.And inter-scale non-cause MRF and intra-scale cause MRF are used to construct the prior model in label field model.Then the Bayes theory is used to segment the image.The experiment results prove that the approach is feasible and applicable.

关 键 词：贾卡经编针织物 图像分割 多分辨率 层次 马尔科夫随机场 

分 类 号：TS181.8[轻工技术与工程—纺织材料与纺织品设计]