Research Of Image Fusion Based On Nonsubsampled Contourlet Transform

Multi-sensor image information processing system is a tendency of the development of information technology and also is the result of emergency of many different advanced image sensors. Image fusion is one of the most important branches of multi-sensor data fusion, which has broad prospects for development as a new research area. It accesses to a more accurate, comprehensive and reliable description of a scene or object by extracting and synthesis of information from multi-sensors. Then we can carry out further analysis of the image and target detection, identification or tracking. It can be predicted that with the continuous development and improvement of the theories, image fusion will have a much wider application in military, remote sensing, robotics, medical image processing and computer vision fields. After nearly three decades of development, image fusion technology has formed a certain scale and many fusion systems were developed. But this does not mean this technology has improved. There are still many theoretical and technical issues to be resolved. In particular, the research work of image fusion in our country began late than the international. So there is an urgent need for extensive basic theory and technology research.The paper focuses on the multi-sensor image fusion technology based on Nonsubsampled Contourlet and its application not only in medical image fusion but also in multi-focus image fusion. Firstly it introduced the concept and benefits of image fusion, and then the methods commonly used in the three levels of image fusion were introduced. Finally, the subjective and objective evaluation criteria of the effect of fusion were summed up.The traditional image fusion algorithms use wavelet transform which is based on multi-resolution. However, the two-dimensional wavelet has poor selectivity of directions and it can't access the outline of the image effectively. Therefore, we can't get a sparse and effective expression of the image. People proposed a series of methods which can capture the geometric structure information of the image effectively such as the Contourlet transform. The Contourlet transform is multi-direction and multi-scale, but it is not shift-invariant. The disadvantage will produce pseudo-Gibbs phenomenon around the singularity which weakening it's frequency-domain locality and direction selectivity. So some people proposed the nonsubsampled Contourlet transform. The image is decomposed by nonsubsampled pyramid and nonsubsampled directional filter banks in a multi-direction way. The nonsubsampled Contourlet transform contains the characteristics of Contourlet transform and also is shift-invariant. The designing of the filter banks is very simple. So it is very suitable for image processing.In this paper, we proposed a new multi-focus image fusion algorithm based on NSCT and PCNN. Pulse coupled neural network is a new neural network which is different from traditional artificial neural network. It can be widely used in image processing, target identification, communication and decision-making optimization and so on. The algorithm achieved good fusion effect by making use of NSCT and PCNN.