研究目的
Investigating the use of an unsupervised approach with Haralick Correlation texture feature for the segmentation of the optic disc in colored fundus retinal images to aid in the efficient diagnosis of glaucoma.
研究成果
The proposed unsupervised method using Haralick Correlation feature achieved high accuracy rates of 98.59% for grayscale and 98.36% for green channel on the DRIVE database, demonstrating effectiveness for optic disc segmentation in glaucoma diagnosis. Future work will focus on evaluation with additional datasets and method improvements.
研究不足
The method was evaluated only on the DRIVE database; future work should extend to other datasets. Potential issues include misclassification due to preprocessing noise removal and disappearing blood vessels in the optic disc region.
1：Experimental Design and Method Selection:
The study uses an unsupervised approach based on Haralick Correlation texture feature for optic disc segmentation. Methods include preprocessing with median filtering and top-hat transform, segmentation using local adaptive thresholds derived from correlation features, and postprocessing with morphological operations.
2：Sample Selection and Data Sources:
The DRIVE database is used, containing 40 retinal fundus images (20 training, 20 testing) captured with a Canon CR5 camera, with spatial resolution of 565 x 584 pixels and 24-bit grayscale resolution.
3：List of Experimental Equipment and Materials:
MATLAB 2014a software for implementation; Canon CR5 camera for image acquisition; median filter with window size n=6; disk-shaped structure element for top-hat transform.
4：Experimental Procedures and Operational Workflow:
Preprocessing involves extracting green channel or converting to grayscale, applying median filter, and top-hat transform. Segmentation computes GLCM and correlation features to derive adaptive thresholds. Postprocessing uses morphological opening and subtraction of image mask.
5：Data Analysis Methods:
Accuracy metric is computed as (TP + TN) / (TP + TN + FP + FN), where TP, TN, FP, FN are true positives, true negatives, false positives, and false negatives, respectively.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
