Principal Component Analysis in Digital Image Processing for Automated Glaucoma Diagnosis

The digital image processing (DIP) techniques usually generate attribute vectors that tend to contain a large number of elements. When working with automated classification techniques, some commonly verified attributes may have low relevance for solving a specific problem or even worse the classification, unnecessarily increasing the dimensionality of the problem. A limited set of relevant attributes simplifies the representation of the image, and consequently, a better interpretation of the data occurs. In this perspective, this research applied Principal Component Analysis (PCA), a widely disseminated technique for reducing dimensionality in the literature, to the attribute vector generated by the DIP, with the aim of increasing the accuracy of the classification of this vector. As a case study, retinal image classification for the diagnosis of glaucoma was used. The data set used was the second version of RIM-ONE, provided by the Medical Image Analysis Group of the University of Laguna, Spain. The results showed that with the application of the PCA, a better classification of the images occurs. With only 7 components, a better classification was obtained than the original data set, which has 36 attributes. These results validate the possibility of applying PCA to optimize the automated glaucoma diagnostic process.