In the past decade, an exciting new theorem, known as Compressive Sensing or Compressed Sensing (CS), mathematically establishes that relatively small number of non-adaptive, linear measurements can harvest all of the information necessary to faithfully reconstruct sparse or compressible signals. This leads to the reduction of sampling rates, storage volume, power consumption and computational complexity in signal and image processing.
The thesis develops three different applications of compressive sensing in multimodal images. The first application presents an effective multi-image fusion scheme based on a Discrete Cosine Transform (DCT) sampling model for compressive sensing imaging by taking advantage of the sparsity of the image in the spectral domain. In the second application, although the depth images delivered by 3D vision system based on Time-of-Flight (ToF) camera provide new perspective, they suffer from relatively low spatial resolution in comparison with color images due to the size limitation of current 2D pixel array in ToF sensor. Hence, the soft solution (i.e., post-processing) to increase the spatial resolution deserves to be advocated in comparison to high payoff of the hard solution (i.e., hardware improvement). From this point of view, the thesis attempts to explore the potential approaches within the framework of compressive sensing to enhance resolution of depth image. Regarding the third application, a rapid development of related research work with regard to processing and analysis of multi-modal image data is urgently desired due to the 2D/3D vision system which not only provides 2D view of the scene but also depth information of the same scene has become increasingly attractive. Hence, the thesis also attempts to explore potential approaches based on CS to sense change in the multi-modal images.
The work presented in the dissertation is expected to contribute to the related field by addressing the following aspects:
• For multi images fusion, in order to reduce the computational complexity and to save storage space, an effective fusion scheme based on compressive sensing is presented.
• To enhance the lateral resolution of depth image, a novel method is proposed by adopting the recent emerging theory of compressive sensing. This approach benefits from the finding of the sparsity of depth image, which is different from the conventional methods and open a new way for super-resolution reconstruction of depth image.
• In the point of view of the sparsity in image analysis, the thesis presents an innovative approach to detect change occurred in multi-modal images. The proposed approach mainly focuses on sparse feature pursuit with arbitrary shape and reconstruct via matrix decomposition. So far, to our knowledge, matrix decomposition has not yet been applied to the multimodal image data. Thus, this formulation yields a novel model for this application.