الفهرس | Only 14 pages are availabe for public view |
Abstract Confocal microscopy imaging is modern technique that is used to image the human eye. Both corneal and retinal images are obtained with this imaging technique. This thesis is mainly concerned with the processing of corneal images. The images can be obtained as separate images or video frames. These images are characterized with hexagonal shapes in Endothelium layer. These images have low contrast. They may contain abnormalities or anomalies. Hence, anomaly detection techniques can be implemented on these images for the purpose of early diagnosis. The main objective of thesis is to perform the anomaly detection task from corneal images, efficiently. Both Machine Learning (ML) and Deep Learning (DL) approaches are introduced in the thesis for anomaly detection with high accuracy. The first approach based on ML adopts Mel-Frequency Cepstral Coefficients (MFCCs) as feature extraction with Support Vector Machine (SVM) classifier. The second approach depends on Deep Learning (DL) concepts. Both Convolutional Neural Networks (CNNs) and Long Short Term Memory (LSTM) are considered in the approach on the frames of corneal videos. Simulation results proved that the utilization of a CNN with 5 layers gives the best classification results for anomaly detection. The results of the proposed techniques will be compared with recent published results for the same data and environments. For more investigation, the percentage of the training and testing will be changed and test the performance. Moreover, a scenario for medical communication is considered with a security framework based on watermarking. |