本文研究了工业生产中对物体表面缺陷进行自动视觉检验中的问题,建立了单个缺陷信号及缺陷过程马尔可夫链模型(考虑白噪声背景)。考察了直接积分、维纳滤波及指数滤波等检测方法。实验表明,对于物体表面缺陷的自动检验是可以实现的。当缺陷窄而密时宜用直接积分法。对其它情形;为简捷计,在对每个象素进行阈值检测前可以先进行指数滤波。 This paper studies the problem of automatic visual inspection of surface defects in industrial production, and establishes a single defect signal and defect process Markov chain model (considering white noise background). The detection methods such as direct integral, Wiener filter and exponential filter were investigated. Experiments show that the automatic inspection of surface defects can be achieved. When the defect is narrow and dense, it is appropriate to use the direct integration method. For other cases; for simplicity, exponential filtering can be performed prior to threshold detection on each pixel.