By considering the causes of image blur in machine vision,the point spread function（PSF） is described with the generalized Gaussian function,and an blurred edge model is established.Then,some samples are obtained from the edges of an image.Moreover,an integration rule is proposed to calculate the PSF of the image according to the estimated results of all the samples.Finally,the optimal PSF and sub-pixel level edge position are estimated via the least squares approximation.Experimental results show that the proposed edge detection algorithm is robust to noise due to the accurate estimation of PSF,and that the dimension error based on the sub-pixel-level edge detection is less than 0.5%.