Automated Edge Detection Technique for Pap Smear Images Using Moving K-Means Clustering and Modified Seed Based Region Growing Algorithm
Nor Ashidi Mat Isa · 2005
In previous studies, conventional seed based region growing (SBRG) has successfully been used to detect the edges of certain regions of interest on digital images. The SBRG algorithm offers several advantages over other conventional edge detection algorithms based on gradient decision; the edges of regions found are perfectly thin and fully connected, and the algorithm is very stable with respect to noise. However, two parameters of the SBRG algorithm, which are threshold value and initial seed point location, must be determined manually. Thus, it is timeconsuming and the edge detection performance is highly subjective to the user. Besides that, the SBRG algorithm cannot avoid trapped seed point problem, which causes incomplete edge detection process. The SBRG algorithm also can only detect the edge of one region of interest in one time. To avoid those problems, the current study proposed an automated edge detection technique. The proposed technique consists of moving k-means clustering and SBRG algorithm. However, the current study modified the SBRG algorithm to enhance its capability in edge detection process. The modified seed based region growing (MSBRG) algorithm is able to detect edges of more than one regions of interest as well as differentiate those edges and can avoid incomplete edge detection process as compared to conventional SBRG algorithm. In the proposed automated edge detection technique, firstly, moving k-means clustering algorithm is used to find the thresholds values automatically. After that, based on the thresholds values, the proposed MSBRG algorithm is applied to detect the edges of regions of interest. Then, the proposed technique is applied to Pap smear images to detect the cytoplasm and nucleus edges of cervical cells. The results obtained show that the proposed automated edge detection technique produce better edge detection performance as compared to conventional SBRG, Cubic Spline, Frei Chen, Kirsch, Laplacian, Prewitt, Roberts, Robinson and Sobel algorithms.