The goal of this project is to study the Principal Components Analysis (PCA) and apply it to ovocyte images. The PCA allows a minimal representation in the least square sense. This allows a massive reduction of the parameters for an image. This is an important step towards image classification.
In our case, we have a collection of ovocyte images we wish to analyze with PCA. To enhance the compactness of the representation by PCA, we preprocess the images and in particular register them.
The processing and analysis presented in this report is aimed at obtaining a parametric representation that enables ovocyte classification into fecund or non-fecund. Based on this representation
scheme obtained from training set of images, the long-term objective is to classify new images to determine if the corresponding ovocyte is fecund or not. The principle interest in such a result is that the law forbids the implantation of more than one ovocyte, thus the need for ensuring it to be fecund.
The interest in preprocessing is obvious for the quality of results obtain with PCA. Nevertheless, some points can be revised and enhanced in the preprocessing. The most interesting area to improve seems to be the ellipse fitting. The ellipse fitting method we used during this project is derived from classical, least square conic section fitting, subject to strong influence from absurd points. Other methods, like the five point fit ellipse fitting methods presented by Rosin , could be investigated and may be useful for cases where parasitic boundaries cannot be eliminated.
To further study the possibility of ovocyte classification, statistical analysis should be done on the coefficients associated to images by the KLT reconstruction. Such an analysis should demonstrate if the coefficient are related to the fecundity of the ovocyte and thus confirm or not that the information about the fecundity is contained in the image.