|
I am interested if there popular algorithms and methods for unsupervised feature extraction of image data that I do not know yet. The types I know are: Principal Component Analysis (PCA), Autoencoder neural networks, Restricted Boltzmann machines (RBM), Deep Belief Networks (DBN). I know that are methods like Kernel PCA, nonlinear Embedding, etc. ... But are this methods wide-spread in feature extraction of image data? |
|
Independent component analysis (ICA) and independent subspace analysis (ISA) is another way to extract good features from images. Andrew Ng group makes good use of this is building big networks. Sparse filtering, also coming from the same group, is also inspired by these methods (combined with sparse coding). |
|
Another nice approach is to use (spherical) K-means for feature learning. Check out Adam Coates's work on this, particularly Learning Feature Representations with K-means (PDF). |
|
I recently started to experiment with "sparse filtering" which works pretty nice. On a related note there is "sparse coding" and for images in particular you could consider a convolution version of these. |
