![]() | Only 14 pages are availabe for public view |
Abstract Ophthalmological imaging is widely used for the diagnosis of eye diseases using high quality images of either the cornea or the retina. Unfortunately, the acquired images have limited quality and resolution. The issue of visual quality and resolution enhancement of corneal images is the main concern of this thesis, in this work a framework is introduced for quality enhancement of corneal images with both histogram equalization and interpolation. Two implementations are presented in this framework, the first implementation comprises histogram equalization first and then interpolation. The second one comprises interpolation first and then histogram equalization. Two types of image interpolation are considered: polynomial interpolation and inverse interpolation: LMMSE, Entropy and regularization. The results of these models are compared to achieve the best quality of corneal images. The best quality of corneal images is obtained with regularized interpolation first and then histogram equalization. The rationale behind this conclusion is that the utilization of regularization theory in interpolation preserve smooth areas and enriches edges. Moreover, we have investigated super resolution of corneal images in this thesis. We compare between super resolution technique’s results and different interpolation techniques on corneal images. Simulation results prove superiority of image super resolution as a tool for obtaining high quality corneal images. PSNR, MSE, Entropy, Spectral Entropy are evaluation metrics which used to measure the performance of the proposed models. Finally, automated diagnosis from corneal images is investigated with convolutional neural networks giving high classification results on the base of the accuracy. |