The “best-fit” line is used as a threshold position, which informs the algorithm whether or not an alternate extraction path should be detected. If the encountered black pixel is below the “best-fit” line, then this pixel, along with all the continuous foreground pixels, are ignored. However, if this black pixel exists on or above the “best-fit” line, this is considered to be the starting point of an overlapping stroke This pixel is called the “turning point”. Commencing from this turning point, a path directed around the overlapping stroke is explored. The algorithm attempts to investigate the right hand side of the turning point. If it is possible to reach the top row of the image, then the extraction path is found.
Otherwise, if the traversal to the right hand side is blocked, then it will go back to the turning point, and traverse towards the left hand side. As shown in Figure 3, both left-hand and right-hand segmentation paths of the character “t” are detected. Once an extraction path is detected, all pixel coordinates are stored for the purpose of character extraction.