I have a dataset composed of $970$ images with size $256\times256$, so I have a data matrix $X \in \mathbb{R}^{970\times65536}$. My idea is to compute the PCA transformation in the training phase considering the whole data-set and then transform the single images before sending them to a classification algorithm (e.g., SVM or a DNN).

The problem is that the eigenvalue decomposition requires is it practically infeasible with a covariance matrix $C \in \mathbb{R}^{256^2\times256^2}$. Is it still possible to use PCA in this situation?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.