Study Of Multiscale Fusion Theory And Its Application

Both the image data fusion method and the wavelet analysis theory almost simultaneously got rapid development in the 1980 s. At present, they are widely used in image processing. One of the very important functions of the image fusion method is to denoise and enhancement an image. With the development and perfection of the wavelet theory, the image fusion technology in multiscale wavelet domain also continues to develop. At present, discrete real wavelet transform and complex wavelet transform are mainly used in image de-noising and image fusion. The quaternion wavelet have a huge advantage over the traditional wavelet in image processing, for it is a new multiscale analysis tool which has a good approximate translation invariant feature and can provide an amplitude information and three phrase information on a different scale for image processing. This paper mainly studies the correlation theory of quaternion wavelet transform and its applications in image de-nosing, as well as its applications in image fusion when combined with non-local average de-noising.Based on the Hilbert transform of two-dimensional signals, the relevant theories and properties of the quaternion as well as the multiscale wavelet analysis theories and methods, Hilbert transform, the quaternion algebra and the wavelet theory method, some properties of quaternion wavelet transform in 2 2L(R) space are studied. In addition, by studying the constitutive properties of the quaternion wavelet transform, this paper constructs an analysis and synthesis filter of quaternion wavelet transform, thus realizing the quaternion wavelet decomposition and reconstruction of a twodimensional image.Based on the traditional wavelet threshold de-noising model, this paper proposed an improved Bayesian threshold. First, we applied threshold de-noising to the quaternion wavelet coefficients modulus value of the image, through the selection and processing of quaternion wavelet coefficients modulus values, we can work out an estimated quaternion wavelet coefficients by the information of modulus and phase, thereby obtaining a restored image. You can hardly see the recover traces of the image in the result to which we applied quaternion wavelet transform threshold de-noising, visual effects and image quality of the image has been greatly improved, the noise has been effectively suppressed as well. Further more, through the analysis and research of the characteristics of quaternion wavelet threshold and non-local means de-noising algorithm, we proposed the image fusion de-noising algorithm based on non-local means and quaternion wavelet threshold。Fusing non-local means de-noised image with quaternion wavelet threshold de-noised image in quaternion wavelet domain, by some certain rules, a fused image will be obtained. The experimental result shows that the quality of fused image obtained from this method is much better than the first two de-noising methods, enhanced the image border as well. In PSNR, this method improved 1.0 to 1.3 as compared with quaternion threshold algorithm, 0.7 to 1.0 with non-local means de-noising algorithm, the visual effect of the image has been greatly improved too.