Abstract:
Retrieval of retinal vascular network is used for diagnosis, treatment, screening, evaluation and
clinical study of many diseases which induce changes in retinal vascular network. Blood vessels
are the predominant and most stable structures appearing in the retina; therefore reliable vessel
extraction is a prerequisite for subsequent retinal image analysis and processing.
This work presents a new approach to extract the vascular tree from monochromatic retinal
images by combining the detection of vessel centerlines with morphological processing.
Vascular skeleton is acquired by detecting vessel centerlines and segmented vascular image is
obtained by a sequence of morphological operations. The vessel centerlines, considered as
local intensity maxima along vessel cross profiles are extracted by the application of directional
differential operators and then evaluation of combination of derivative signs and average
derivative values. Two separate morphological image processing methodologies are exploited
for vessel segmentation. In multi scale morphological reconstruction, the vessels are enhanced
by applying a modified top hat operator with variable size circular structuring elements aiming
at enhancement of vessels with different widths. The binary maps of the vessels are obtained
at four scales by using morphological reconstruction with double threshold operator. A final
image with the segmented vessels is obtained by iterative seeded region growing process of the
centerline image with the set of four binary maps. In morphological bit plane slicing, a
multidirectional top hat operator with rotating structuring elements is adapted with the
Gaussian-like profile of vessel and later bit plane slicing is used to extract visual information of
vascular network. A region growing method is applied to integrate the centerline and the
images resulting from bit plane slicing of vessel direction dependent morphological filters.
These methods are evaluated using the images of two publicly available databases, the DRIVE
database and the STARE database. The results of this work are compared with those from
other recent methods, leading to the conclusion that our algorithm is comparable with other
solutions, while approximating the average accuracy of a human observer without a significant
degradation of sensitivity and specificity, with significant improvement in processing time.
