Fig 1.
The proposed model.
Fig 2.
The first line represents the RGB images and the second row represents their images converted to grayscale.
Fig 3.
Output of morphological tactics on each channels(Red, Green and Blue).
Fig 4.
Homomorphic filter output on each channel.
Fig 5.
Output of Wiener filter on each RGB channel.
Fig 6.
Output of the 2nd order multi-dimensional Gaussian derivative identifier, where (a) denotes the primary output with the maximum value, and (b) represents the normalized factor applied to the initial output.
Fig 7.
Oriented Diffusion Output, where (a) represents the retinal image prior to applying the Coherent Anisotropic Diffusion filtering operation, and (b) represents the proposed scheme of Coherence Anisotropic Diffusion filter.
Fig 8.
The histogram identified two vertical bars to serve as the two thresholds.
TU is obtained by subtracting 0.7 times the standard deviation from the mean value of the image histogram, while TL is obtained using the mean value of the edge-based histogram.
Fig 9.
Displays the various images obtained during the segmentation process.
The mask image and marker image are presented in Fig(a) and (b), respectively. The morphologically reconstructed image is depicted in Fig (c), while the finilized binary image of segmented vessels is illustrated in Fig (d).
Table 1.
Analysis of segmentation method’s performance.
Fig 10.
(a) Segment Image without Enhancement (Green lines indicates the missing Vessels (b) Segment Image with Enhancement.
Note: Green lines indicates the missing Vessels.
Fig 11.
Our proposed method was evaluated on images with pathologies, and the results are presented in Fig (a and e), which depict the ground truth images. Fig (b and f) show the results obtained by the Nuygen method on these images, while Fig (c and g) show the results obtained by the Hou method. Fig (d and h) shows the results obtained by our proposed method on these images.
Table 2.
Challenging images performance assessment.
Fig 12.
Receiver Operating Characteristic (ROC) curves for classification of vessels pixels on the DRIVE and STARE databases.
In this Fig, the ROC curves indicate the performance of the proposed algorithm on the DRIVE (Fig(a)) and STARE (Fig(b)) databases.
Table 3.
Impact of 2nd order multi-dimensional LoG Detector on segmentation of retinal vessels.
Table 4.
Segmentation model performance assessment.
Fig 13.
Comparative study of the results obtained by our proposed method and those of Nguyen et al. [49], Hou [50], and Zhao et al. [51] is presented.
Fig (a) shows the result obtained by Nguyen, while Fig (b) shows the result obtained by Hou. Fig (c) and (d) show the results obtained by FR and IUWT-based Zhao, respectively. Fig (e) presents the results obtained by our proposed method, and Fig (h) shows the ground truth image.
Table 5.
Comparison of proposed method with existing methods.