HEP-2 CELL IMAGES FEATURE EXTRACTION BASED ON TEXTURAL AND STATISTICAL ANALYSIS

This project is about Human Epithelial type 2 (HEp-2) Cell Images Feature Extraction Based on Textural and Statistical Analysis. The medical industries have yet to found any reliable solution in differentiating the Anti-Nuclear Antibodies disease according to its cell pattern. Current practice, subject to physician's expertise, is not very reliable and cannot be reproduced. The main objective of this project is to provide significant differentiable features based on textural and statistical features of the HEp-2 cell images. The textural features are basically based on the surface of the cells which are analyzed from the grayscale images of the cells. The features are later classified to test its reliability. In this project the images will be analyzed in grayscale mode and processes using two different order of statistical analysis. The second order statistical analysis contains the textural features representation. It was found out that homogeneity and correlation of patterns are the same. Hence, avoid using this feature in order not to have wrong classification information. Also not all Gray-Level Co-occurrence Matrices (GLCM) properties features can be used to differentiate HEp-2 cell patterns. At the end of this project, the results shows that the use of textural (second order statistical) analysis is beneficial to get better accuracy of classification, though it still depends on the type of classifier used.