Research Of Image Fusion Based On Nonsubsampled Contourlet Transform And Pulse Coupled Neural Network

Image fusion is a new technology that combines many other modern high-technologies, such as signal processing, artificial intelligence, image processing, computer technology, displaying and sensors. In order to describe the scene images completely and exactly, image fusion combines multiple images on the scene from the different sensors, and finally obtains a new image on the scene. Image fusion plays an important role in geographic information systems, medicine, military fields, industrial, surveying and intelligent robots.In this dissertation, in order to enhance the application of image fusion, we studied a new algorithm of image fusion based on pulse coupled neural networks (PCNN) and multi-scale decomposition. The main research contents can be summarized as follows:1. The background, significance, and research methods of image fusion are introduced in this dissertation. The category of image fusion and the theory about PCNN as well as nonsubsampled contourlet transform (NSCT) are described.2. The ordinary algorithm of image fusion and algorithm of image fusion based on PCNN are compared in the experiment.And the paper analyzes the characteristics and performance of PCNN.3. The fusion results of various tools of multi-scale decomposition and reconstruction are compared in the experiment, and the experimental results show that the algorithm of image fusion based on NSCT is much better than the method of mean value, wavelet transform and contourlet transform.4. The adaptivity of the coefficients selection principles of traditional algorithm of image fusion based on PCNN and multi-scale decomposition is not considered, and it only selects the gray level of pixels as the external input of PCNN neurons. The human visual system is sensitive to the edge detail of the image information, so if we only apply the above method, we can’t obtain good visual characteristics. In order to overcome the drawbacks, an algorithm that selects spatial frequency as linking strength of PCNN are proposed, the linking strength of PCNN can be selected adaptively,so that, the fusion rules can adapt the characteristics of the image much better.5. A novel algorithm of image fusion based on PCNN and NSCT is proposed. In order to improve the efficiency and save more time, the high-frequency subband and the low-frequency subband don’t choose PCNN at the same time. Because a high clarity of the image is required in the low-frequency subband, a novel algorithm based on edge and region energy and local clarity is proposed in this dissertation. For the high-frequency subband, because the highest subband is the edge information, all of it must be reserved, so selecting large absolute of coefficients is employed in the highest subband. Selecting spatial frequency as linking coefficients is employed in the other subband, so the visual characteristics of the fusion result are enhanced.6. The validity of the proposed fusion rules is demonstrated by fusion experiment. The fusion results show that the fusion result of the algorithm proposed in this dissertation is much better than traditional algorithm in terms of both objective evaluation criteria and visual quality, such as common algorithms based on wavelet transform, contourlet transform and NSCT.