Image Fusion Research Based On Dual-Tree M-band Wavelet Transform

Because of the limitation of objectivity when the optical sensors image in one area, it is difficult to describe every detail of a scene. This problem can be solved by image fusion which has been becoming a hot spot in the area of image processing. It is widely used in these areas such as medical diagnosis, machine vision, and object recognition and so on. This paper apply dual tree M-band wavelet(DTT) on image fusion at the first time, because of its much lower shift sensitivity and better directionality, it has made good effects.This paper focuses on the image fusion based on dual-tree M-band wavelet transform, making in-depth discussion and research on the existing algorithms from the following two aspects, the dual-tree M band wavelet transformation and the rules of image fusion, and compares it with other algorithms. Pixel fusion algorithm based on multiresolution is quite close to the signal processing mode of people's ears. So the fusion algorithm based on multiresolution such as discrete wavelet transforms (DWT) makes good effects. And DWT becomes the main algorithm. But DWT has some drawbacks. The drawbacks are shift sensitivity and poor directional properties. Dull-tree M-band wavelet can vary the two drawbacks. Dual-tree M-band wavelet take two classic wavelet tree in parallel, with ?lters forming approximate Hilbert pairs, through an approximate Hilbert pair to formulate the"dual"wavelets. It obtains a much lower shift sensitivity and better directionality in two dimensions than the DWT. When related to fusion rules, we use Pulse-Coupled Neural Networks (PCNN) to process low frequency coefficients and use improved PCNN to process high frequency coefficients. It has been demonstrated that the rules we use is better that the other fusion rules by our experiments. The experiment results indicate that, whether it is based on average value, root-mean-square error, average gradient, entropy, or subjective visual evaluation, the proposed selection methods are feasible and effective.