Exercise in Computer Vision A Comparison of Thresholding Methods – NTNU –

This paper contains a comparison of common, simple thresholding methods. Basic thresholding, two-band thresholding, optimal thresholding (Calvard Riddler), adaptive thresholding, and p-tile thresholding is compared. The different thresholding methods has been implemented in the programming language c, using the image analysis library Xite. The program sources should accompany this paper. 1 Methods of thresholding Basic thresholding. Basic thresholding is done by visiting each pixel site in the image, and set the pixel to maximum value if its value is above or equal to a given threshold value and to the minimum value if the threshold value is below the pixels value. Basic thresholding is often used as a step in other thresholding algorithms. Implemented by the function threshold in thresholding.h Band thresholding. Band thresholding is similar to basic thresholding, but has two threshold values, and set the pixel site to maximum value if the pixels intensity value is between or at the threshold values, else it it set to minimum. Implemented by the function bandthresholding2 in thresholding.h P-tile thresholding. P-tile is a method for choosing the threshold value to input to the “basic thresholding” algorithm. P-tile means “Percentile”, and the threshold is chosen to be the intensity value where the cumulative sum of pixel intensities is closest to the percentile. Implemented by the function ptileThreshold in thresholding.h Optimal thresholding. Optimal thresholding selects a threshold value that is statistically optimal, based on the contents of the image. Algorithm, due to Calvard and Riddler: http://www.ifi.uio.no/forskning/grupper/dsb/Programvare/Xite/