Multiscale analysis of diagnostic features from color fundus images
No Thumbnail Available
This thesis work is an investigation of multiscale analysis of diagnostic features of the color fundus images. Analysis of diagnostic features is done in the multiscale domain by wavelet transform. There are three major contributions. First, the different wavelet subbands are investigated for the presence of retinal features such as exudates and microaneurysms. Secondly, the pathological features are extracted. Blood vessels are extracted by using Kirsch and Gaussian matched filter (MF) based method. Inpainting is performed on the green channel image by considering the blood vessels structure as a mask for detecting the amount of blood vessels affected by pathologies. Independent component analysis (ICA) is used in multiscale for extraction of exudates and detection of changes in the retina, which occurs due to the progression of diabetes. Third, glaucoma is detected by evaluation of differential entropy in wavelet subband. The performance of different methods is analyzed by different quantitative measures such as, specificity, sensibility, accuracy, universal quality index (UQI), Pearson correlation coefficient (PCC), and structural similarity (SSIM) etc.
Supervisor: Samarendra Dandapat
ELECTRONICS AND ELECTRICAL ENGINEERING