Objective:
The objective is to provide a precise segmentation technique based on ACRF which can handle the variations between major and minor vessels and reduces the interference present in the model due to over fitting and can provide a high-quality reconstructed image. Therefore, a robust method with statistical properties needs to be presented to enhance the performance of the model. Moreover, a statistical framework is required to classify images precisely.
Methods:
Adaptive Conditional Random Field (ACRF) model to detect DR disease in early stages. Here, major vessel potentials and minor vessel potential features are extracted which in precise segmentation of vessel and non-vessel regions. This feature enhances the efficiency of the model. These major vessel and minor vessel potential features rebuild the retinal vasculature parts precisely and help to capture the contextual information present in the ground truth and label images. This method utilizes an ACRF model to reduce interference and computation complexity. Here, two efficient features are extracted to segment fundus images efficiently such as major vessel potentials and minor vessel potentials. The proposed ACRF model can provide the design patterns for both input images and labels with the help of major vessel potentials, unlike state-of-art-techniques which provide patterns for only labels and model the contextual information only in labels which is very essential while performing vessel segmentation
Results:
The performance results are tested on the DRIVE dataset. Experimental results verify the superiority of the proposed vessel segmentation technique based on the ACRF model in terms of accuracy, sensitivity, specificity, and F1measure and segmentation quality.
Conclusion:
A highly efficient vessel segmentation technique is evaluated to describe major and minor vessel regions efficiently based on the ACRF to recognize DR in early stages and to ensure an effective diagnosis using eye fundus images. The segmentation process decomposes input images into RGB components through histogram labels based on the proposed ACRF model. Here, the Gabor filtering approach is used for pre-processing and predicting parameters. The proposed segmentation method can provide the smooth boundaries of minor and major vessel regions. The proposed ACRF model can provide the design patterns for both input images and labels with the help of major vessel potentials, unlike state-of-art-techniques which provide patterns for only labels and model the contextual information only in labels.
EFFICIENT SEMANTIC SEGMENTATION OF MAN-MADE SCENES USING FULLY-CONNECTED CONDITIONAL RANDOM FIELD