Analysis of a 3D recursive filtering edge detector for extracting muscle fiber orientation

We implement two recursive filtering and hysteresis thresholding algorithms. The first algorithm (RHT) is proposed by Monga and Deriche (1991) and the second algorithm is a recursive filtering and hysteresis multilevel hysteresis thresholding algorithm (RHMT). The algorithms are implemented on 3D images of thigh muscles. The RHT method is effective and involves fewer calculations than many other algorithms, but the choice of thresholds poses a problem. To get the most optimal thresholds, the hysteresis thresholding has to be run a few times. The RHT method has been modified to a multilevel hysteresis thresholding method in which the thresholding is done iteratively over a range of thresholds. The result is a group of multiple thresholded edge images, which take lesser time compared to multiple runs of the RHT algorithm. The average of the resultant output gradient images of the iterations of the RHMT algorithm can be taken as the most effective edge detected image. Results of this method and the original method on some 3D images are presented