Tesis doctoral de Julio César Rolon Garrido
This thesis investigates the use of the generalized lifting method for the sparse representation of images, with application to lossy compression. the generalized lifting method is a signal decomposition scheme with perfect reconstruction that enables the use of non-linear operators. In this thesis, the application of the method is conducted in the wavelet domain, with the objective of increasing the sparsity of the wavelet signal through the use of a non-linear, invertible, context-based, generalized lifting mapping operation. The idea of using the generalized lifting method in the wavelet domain is similar to those of the adaptive directional lifting methods and bandelets. three approximations to the generalized lifting decomposition are studied: in the rst case, in which the image to be coded belongs to a class; the pdf is estimated for the class of images, and this statistical model is used to decompose the image. the second approach is conceived for generic images. The idea is to perform a global estimation of the pdf of the image, and to use this estimation for the generalized lifting decomposition. Most of the energy in the wavelet subbands is located along the contours of the image; therefore, a set of context-based, contrast-invariant models of contours has been devised as the realization of the pdf that is used to decompose the image. the third solution is a local adaptive pdf estimation approach, also devised for generic images, in which a simplied pdf model is used to grasp the local dynamics of the image subbands. the generalized lifting signal is encoded with ebcot, a context-based, bit-plane entropy coder that is the core of the jpeg2000 standard. The changes in the statistics of the wavelet signal produced by the eect of the generalized lifting mapping, and their consequences for the performance of the entropy coder are also investigated.
Datos académicos de la tesis doctoral «Generalized lifting for sparse image representation and coding«
- Título de la tesis: Generalized lifting for sparse image representation and coding
- Autor: Julio César Rolon Garrido
- Universidad: Politécnica de catalunya
- Fecha de lectura de la tesis: 25/01/2010
Dirección y tribunal
- Director de la tesis
- Philippe Salembier Clairon
- Tribunal
- Presidente del tribunal: ferran Marqués acosta
- joel Sole rojals (vocal)
- gemma Piella fenoy (vocal)
- Luis Salgado álvarez de sotomayor (vocal)