In this paper, different image preprocessing methods are compared based on their ability to remove noise and to segment the images. Two filters, namely, Median filter and Wiener filter, and seven image segmentation methods, namely, Sobel, Prewitt, Roberts, Laplacian of Gaussian (LoG), Canny edge detection, basic global thresholding (BGT) and Otsu's global thresholding (OGT) are applied on the images of eight different signal-to-noise ratios (SNRs) ranging from 2.7 to 17.8. First, the preprocessing results are qualitatively compared by visual inspection for image SNRs of 17.8 (high) and 2.7 (low). Then the effects of different preprocessing methods are quantitatively analyzed by determining the accuracy of centroid detection of circular marks using Hough transform. The quantitative comparison showed that Median filter plus BGT or OGT give better results than other methods for low SNRs, and Wiener filter plus LoG detector provided higher accuracy compared to other methods for high SNRs. The application of this work is in many areas, for example, biomedical imaging, flow diagnostics and computer vision, where we detect the sizes or locations of circular objects in images.