Tesis doctoral de Jorge González Conejero
This work is focused on coding digital images that contain areas without useful information, here named as no-data regions. In many different fields, images contain regions that should not be considered, and ignoring the lack of importance of these no-data regions could penalize the compression performance and the later post-processing applications. In this work, we analyze the changes in the behavior of different stages in coding systems that the arising of no-data regions produce. Taking into account that the discrete wavelet transform (dwt) stage used in most coding systems is devised to decorrelate the information of the image detecting smooth areas and sharp edges, one can easily infer that the most penalized stage is the dwt due to the boundary between the data and no-data regions. Also, the inclusion of no-data coefficients in the fractional bitplane encoder (bpe) increases the final bit budget with useless information. in the literature there are coding systems that are able to encode only certain regions within the image, making them suitable to deal with no-data regions. For instance, bisk is a coding system that is specifically devised to handle no-data regions. Nevertheless, some applications have the requirement of providing a standard encoded image. Jpeg2000 is one of the latest standards developed by the joint photographic experts group (jpeg), addressing the manipulation, security, and transmission of still images and video. Jpeg2000 fulfills most of the requirements from applications in many different scenarios. Nevertheless, the jpeg2000 standard does not consider to encode only certain regions within the image. We propose five techniques for mono-component images within the jpeg2000 framework, the extension to a multi-component scenario of the best proposed technique, and a coding scheme based on bisk and jpeg2000, that could be generalized to any shape-adaptive (sa) coding system, for multi-component images containing no-data regions. The main goal of these proposals is to minimize, or even eliminate, the cost of coding regions without information. for mono-component images, the first two techniques are carried out as simple pre-processings that modify the no-data samples, average data region (adr) and phagocyte. The aim is to smooth the boundary between the data and no-data regions. The third technique is based on the maxshift roi coding method provided by the jpeg2000 standard. Maxshift allows to prioritize data ahead of no-data regions. The last two techniques imply modifications to the jpeg2000 core coding system, the dwt-sa applies the sa version of the dwt, and the bpe-sa (also called sa-jpeg2000), in addition to the shape-adaptive dwt, implements a fractional bpe to avoid the no-data coefficients in the whole encoding process. These last two techniques do not provide a compliant bitstream. Experimental results suggest that the best technique among the proposed is the bpe-sa, however, it does not provide a compliant bitstream. The best compliant technique varies in function of the image data set, although adr is the one that requires the lowest computational load. in remote sensing, most sensors capture images containing multiple components. One important feature is the redundant information across the components. The karhunen-loíªve transform (klt) and the dwt are applied to exploit the spectral redundancies. Also, the best proposed technique, sa-jpeg2000, is extended to deal with multi-component images with no-data regions. The result is a multi-component sa-jpeg2000 that provides superior coding performance than state-of-the-art sa coding systems as: sa-wdr, sa-tarp, and bisk. there are two main strategies to encode multi-component images that contain no-data regions. The first one applies a spectral and spatial decorrelation separately, and then a two-dimensional (2d) sa coding system. The second strategy applies a three-dimensional (3d) transform and a 3d bpe. We propose a scheme that is able to apply both strategies, it is based on klt and dwt to decorrelate the image, and either on 2d-bisk, or 3d-bisk, or sa-jpeg2000 to encode the image. Results obtained suggest that the first strategy provides the best coding performance. Results also suggest that the best spectral transform is the klt, and that sa-jpeg2000 provides similar coding performance than bisk, however, jpeg2000 framework considers many features and capabilities to manage digital images.
Datos académicos de la tesis doctoral «Coding techniques addressed to images containing no-data regions within jpeg2000 framework«
- Título de la tesis: Coding techniques addressed to images containing no-data regions within jpeg2000 framework
- Autor: Jorge González Conejero
- Universidad: Autónoma de barcelona
- Fecha de lectura de la tesis: 10/12/2009
Dirección y tribunal
- Director de la tesis
- Joan Serra Sagrista
- Tribunal
- Presidente del tribunal: josep Rifa coma
- david Megías jiménez (vocal)
- (vocal)
- (vocal)