Robust Image Segmentation Using Modified Edge-Following Scheme with Automatically-Determined Thresholds

Abstract

The modified edge-following scheme using automatically-determined thresholds is proposed in this work for the applications of robust image segmentation. Prior to executing the edge-following scheme, global and local threshold values are first determined from the image contents. Analyzing the image characteristics would derive at the global threshold value, which is also used to determine the initial points. After the image frame is partitioned into many small blocks, an average value is computed for the gray-level differences between each pixel and its 8 neighboring pixels. A histogram comprised from these average values is then used to compute the local threshold value, which also is the determining criterion to see if the search procedure should continue or not. By determining threshold values automatically, the scheme proposed in this work improves upon the existing flaws of the conventional schemes for different image types, which in turn derives at robust and good segmentation results.