Local edge detection of an image using weighted pixels based on their neighbourhood

Abstract

In this paper, a new method of edge detection in gray scale images based on comparing a pixel's value to its eight neighbors' values is presented. Each pixel gets a new value based on such comparison. Similar pixels or smooth surfaces are then those which have a zero value resulting from the sum of all values and the edges are the set of pixels whose values are different from zero. This decision of assigning zero to a pixel is based on the use of an automatic range, being the mean, detected from the histogram of differences of all pixels' values. This leads to smoothing the pixels of the same regions and keep the contours only. The proposed method gives visually very similar results to some famous edge detection algorithms but with less complexity. The results can be used into to two forms in order to accommodate the objective usage.