Research On Image Fusion Based On Nonsubsampled Contourlet Transform

In order to obtain more accurate, complete and reliable information description about the same object, image fusion can provide complementary information, enrich the amount of information, and improve the adaptability of the environment by using imaging sensors of different imaging modalities. Image fusion technology can overcome the limitations of signal sensor imaging in geometry and spectrum, enhance image credibility and intelligibility, improve the ability of target recognition and data classification, which makes the fused image contain important detail information of source images, and lead to a good result to human visual perception or post-processing for computer. As the result, image fusion has turned into the important branches of information fusion, and has become one of the popular research domain recently, and has a wider application in computer vision, medical image processing, remote sensing and military. After several decades development, there are some achievements of image fusion, which are focus on pixel-level image fusion recently. But image fusion algorithm and image fusion quality assessment are also two kinds of problems in image fusion domain, domestic and foreign researchers mainly focus on researching the two problems which existed in image fusion.With good localized time domain and frequency domain and multi-resolution nature, wavelet has become a mainstream technology in the image fusion application. Recently, due to its limited directional selectivety and anisotropy, the non-subsampled Contourlet transform is proposed. In this paper, we focus on image fusion based on nonsubsampled Contourlet transform, the main contributions are:1. Image edge feature is researched in the nonsubsampled Contourlet domain, we propose a novel method of extracting image edge feature. On the basis, image fusion algorithm based on edge feature is proposed, the experimental results on various fusion algorithms results verify that our fusion algorithm is simple, our algorithm not only improves fusion effects, but also cuts down the time complexity.2. We propose an image fusion algorithm based on pulse coupled neural network (PCNN), which combining image edge feature and PCNN. In the method, edge feature is inputted to PCNN to inspire neurons firing, judging the clear part of source images according to firing time, thus, selecting the fusion coefficients. The experimental results on various fusion algorithms show our algorithm can overcome the limitation of conventional algorithms based on PCNN which use pixel to inspire neurons, the fusion effects has been improved greatly.3. On the other hand, we research image fusion quality assessment methods and propose an image fusion quality assessment method based on image definition and structure similarity. This method takes full account of structural similarity and the characteristic of human visual system, gives different weights to those salience regions by way of definition weight, it does not need the reference image, and provides reliable support for selecting appropriate algorithm in different applications. Experimental results on different image fusion algorithms show that our method has high consistency with subjective evaluation, at the same time, objective evaluation on fusion algorithms of image fusion based on edge features and image fusion based on edge feature and PCNN validates the correctness of our proposed evaluation method further.