Retinal Blood Vessel Segmentation Based On The Adaptive Threshold And Simplification Of Modular Supervised Method
Blood vessel segmentation has been the important issue on medical field. One of methods has been developed
which is Modular Supervised method. Modular Supervised method used normalized Eigen value to characterize object.
However, the results of Gaussian convolution can be determined the threshold value to separated between blood vessel and
background without the Eigen value. We proposed new approach to segment blood vessel on fundus by using combination the
adaptive threshold and simplification of Modular Supervised. Our proposed method takes less computation time than
Modular Supervised method. It is caused, our proposed method ignored two processes on Modular Supervised, which are
determining and normalizing of Eigen value. Our proposed method is started by reading retinal image input and followed by
taking of green channel image. The second stage, green channel image is enhanced by using contrast-limited adaptive
histogram equalization and image convolution by using Laplacian of Gaussian. The third stage, convert to binary image based
on the threshold value of image enhancement results. We use Otsu’s method to determine the adaptive threshold value. To
improve the binary image results, it is necessary to remove the noises using morphology dilation. The last stage, remove the
object border using overlapping mask. Our experimental results show that the lowest, the average and the highest of accuracy
are 84.36%, 88.80% and 91.57% respectively.