Research On Quaternion Wavelet Transform Based Image Fusion

By some kind of algorithm, image fusion makes full use of the information in two or more images and synthesizes them into a fusion image, which makes the information more abundant and reliable. As a new research field in information fusion, image fusion has achieved some achievements in thirty years, and has been applied in many fields, such as medicine, civil, military and other fields. Unfortunately, due to the late start of domestic research, research on image fusion is not deep enough and systematization. In the case of a complete image fusion theory system has not yet formed, its in-depth research still has a very important significance.At present, the research on image fusion algorithm based on the frequency domain is very active, such as the pyramid transform, wavelet transform, and all kinds of multi-scale transform. Among them, quaternion wavelet transform is a novel image analysis tool in multi-scale transform field, which can not only overcome the disadvantage of lacking of shift invariance in wavelet transform and contourlet transform, but also have stronger feature description ability of the image edges, contours and directional texture. More importantly, the unique amplitude and phase of coefficients obtained by quaternion wavelet transform has a 1:4 redundancy, more details and features can be extracted from the source image. Therefore, it is very necessary and valuable to study the quaternion wavelet transform.The paper studies the quaternion wavelet transform after studying the traditional multi-scale analysis tools and the corresponding fusion method. On the basis of this work, the paper further explores the application of quaternion wavelet transform in image definition evaluation and image fusion. The main research contents of this paper are as follows:(1) Through the research of theory and application of the popular multis-cale analysis tools, the paper analyzes the similarities and differences between DWT coefficients and QWT coefficients as well as the significance of those coefficients, and describes the physical meaning of amplitude and phase of the OWT coefficient which further confirme the advantages of QWT.(2) On the basis of QWT, a new image definition evaluation index which is based on the amplitude and phase of the QWT coefficients is proposed.The paper constructs a new image definition evaluation algorithm, which is based on the saying that clear images process clear texture feature. Firstly, the paper uses QWT to decompose an image into low frequency sub-band and high frequency sub-bands, then extracts the corresponding amplitude and phase from the low frequency sub-band and high-frequency sub-bands respectively, then calculates the gradient of amplitude of low frequency sub-band in three directions(vertical, horizontal and diagonal), and uses the obtained gradients to multiply corresponding pixel of phase of low frequency sub-band or the lowest layer of high frequency sub-band, after taking the absolute value and accumulation, two index values obtained.(3) A new image fusion method based on QWT is proposed.First of all, the paper uses QWT to decompose an image into low frequency sub-band coefficients and high frequency sub-band coefficients, then fusion coefficients are obtained by choosing max absolute value coefficients in low frequency sub-band and contextual model in high frequency sub-bands, finally, fusion image is obtained by inverse QWT from the fusion coefficients. The experimental results show that this algorithm retains more details and edge information in source images, and it has a better performance in visual evaluation and objective evaluation compared with the existing fusion algorithm.