| Compressed sensing (referred to as CS) theory is a new coding theory, broken to the Nyquist sampling theorem, which required for keeping the integrity of the signal, must be not less than2times to sampling. If only the signal is sparse or compressed, we can restore the original signal through far less than2times sample rate. Based on the tremendous advantages of compressed sensing theory, at present, compressed sensing theory has been well applied the field of signal processing and image processing, set off for the compressed sensing theory and its application in various areas of research boom, such as a signal of the sparse representation, observation matrix design, recovery algorithm design, object detection, face recognition, image fusion. This paper is to study the recovery algorithm of compressed sensing and image fusion under the framework of compressed sensing, mainly as follows:Image is a complex signal with geometric structure, each pixel is not isolated, and they have certain connection with each other, the general recovery algorithm of compressed sensing rarely take into account the specific structure of the image information (the edge of the texture information, etc.). Full account of this algorithm to the image edge information and the smooth regional differences, for the image details, edge detection introduced the idea, through the Wavelet and Curvelet edge detection technology to locate edge coefficients and enhancing reconstruction results of the image detail information; for the outline of the image information, combined with different coefficients of Fourier domain reflects the different structures of image, designed observation model with the fixed-density and variable density:the central part is whole sampling, the surrounding high-frequency coefficients according to the distance from the center of the density of different random sampling, the more off-center, the smaller the sampling density and sampling points less. By experiment shows that the algorithm regardless of the time complexity and accuracy than the CWSpB reconstruction algorithm has been greatly improved, and compared to BP, MP algorithm, at lower sampling rates is still relatively good reconstruction effect.Described the difference and difficult between the image fusion under compressed sensing and the traditional:the traditional image fusion operated atomic (pixels, transform domain coefficients, etc.) has a direct correspondence to the image itself, and directly reflect the characteristics of the image, while under the CS framework, the observation vector and the original image is no direct correspondence, how to fuse the observation vectors is hot and difficult under the CS framework of image fusion. With the observation model, which proposed in the CS reconstruction algorithm based on image structure model, and the feature of the local Fourier harmonic coefficients fusion, presented a multi-strategy image fusion method under the CS framework. Experiments show that the algorithm has good fusion results. |
