K. Sheng, K. Kameyama, K. Katagishi, and K. Toraichi (Japan)
Fundus image, Fluency analysis, Blood-vessel extraction, Matched filter, Quadratic Fluency sampling function
Matched filter is widely utilized by many fundus blood vessel extraction methods and is mainly designed with Gaussian function based on a cross section model of blood vessel intensity profile. However, from the fact that the bor der of a blood vessel naturally exists, it will be inevitably influenced in the accuracy if a function with infinitely long trails is utilized to match it. In this research, we suppose a refined model for the cross section of blood vessel in tensity profile. Based on this model, Fluency analysis is imported to select matched filter kernels. Five other curves are selected to the matched filter design besides the Gaus sian curve. The matched filter responses (MFR) of all six curves are evaluated with standard blood-vessel extraction images provided by two blood-vessel extraction databases through the computation of true position rate (TPR), false positive rate (FPR) and the extraction error rate. The ex perimental results prove the effectiveness of matched filter based on the new model.
Important Links:
Go Back