Non-linear image representation based on IDP with NN
Roumen K. Kountchev, Stuart H. Rubin, Mariofanna G. Milanova, Vladimir Todorov, Roumiana A. Kountcheva · WSEAS Transactions on Signal Processing archive · 2009
In this paper is offered a method for non-linear still image representation based on pyramidal decomposition with a neural network. This approach is developed by analogy with the hypothesis for the way humans do image recognition using consecutive approximations with increasing similarity. A hierarchical decomposition, named Inverse Difference Pyramid (IDP), is used for the image representation. The approximations in the consecutive decomposition layers are represented by the neurons in the hidden layers of the neural networks (NN). This approach ensures efficient description of the processed images and as a result -- a high compression ratio. This new way for image representation is suitable for various applications (efficient compression, multi-layer search in image databases, etc.).