Research On Compressed Sensing Based Image Fusion With Multiple And Blocking Measurement

Compressive sensing technology becomes a research highlight in recent years with itslow sampling rate and more efficient data reconstruction algorithms. Based on compressedsensing theory, Compressive Imaging (CI) is also gradually advanced. Here this thesis focuseson the fusion of CS based image measurements. Compared with the conventional imagefusion, CS based image fusion is more efficient with lower hardware cost and higherreal-time performance. Aiming at improving fusion rules, compressive measurement fusionmethods in current CS based image fusion system, this thesis explores several solutions. Thecontributions are as follows:1. Two new CS based image fusion rules are proposed after the study of observationmatrix and measurements. The first method, suggests a spacial frequency weighted fusion.fusion of three types of image have obtained good visual perception and most edge preserved.with respect to the sampling operations in star-shaped sampling operator in2D Fourierdomain. The second method, proposes a gradient weighted fusion rule capturing more mutualinformation and edge information with sharp visual sensing.2. Propose using a mixed observation matrix from partial DCT matrix and randomnoiselets matrix to CS based image fusion, as the star-shaped sampling operator has goodability of collecting information but with bad adaptation. The mixed observation matriximproves the performance of conventional single sampling matrix when the sapling rate islow and collected information is not enough. The results of experiments show that, ourscheme get better fusion quality under low sampling rate, and make the fusion image get moresource information.3. In order to get measurements containing image local information, propose using blockdiagonal matrix to image fusion, then design different fusion rules for different measurements.compared with the blocking scheme under structurally random matrix and the conventionalunity observation then fusion method, our proposed scheme get clearer fusion images withmore fine detail such as edge and texture.