Sub-pixel localization of edges in JPEG images

Many methods for edge detection have been proposed in the last two-three decades. Sometimes is desirable to increase resolution of system for edge detection. There are two ways how to deal with this problem. Either a sensor with a greater number of pixels is used or method for edge detection with sub-pixel accuracy is applied. In some cases, only JPEG pictures are used as incoming data. In this paper we analyze how JPEG compression of picture changes the accuracy of edge localization. JPEG pictures with different compression quality were used to test three methods for edge localization in 1-D images with sub-pixel accuracy: method that approximates real 1-D picture with function erf (EFA) and edge detectors based on statistical moments (STM) or spatial moments (SPM) of the image function. This paper presents results of simulations of noiseless pictures and pictures with Gaussian noise. Lower and upper 5% quantiles are chosen as the accuracy criterion.