Research Of Improved Image Fusion Algorithms

With the continuous development of image sensor technology, image fusion has become an important technology in the image processing and machine vision. Multiple images will be processed according to some particular algorithm to obtain the fused image so that it has a richer, more accurate and perfect information by using complementarity. Now image fusion is widely used in target detection, machine vision, medical diagnosis, military reconnaissance and other fields.Multi-resolution image fusion method which is an important part of pixel-level image fusion has been popularly developed and used. To solve the problem which edge information and detail content are easily lost in the process of image fusion, this paper proposes two improved image fusion algorithm from two different angles about wavelet transform and NSCT by analyzing and researching multiresolution decomposition technique to improve the quality of fused image. The main topics in this thesis are as follows:1) Study the characteristics of high and low frequency coefficients decomposed by wavelet transform, then propose an improved image fusion algorithm based on DWT and mathematical morphology, use the fusion method based on the morphological edge detection to handle with low frequency coefficients and adopt improved regional energy method to fuse high frequency coefficients.2) Introduce the relevant knowledge about Pulse Coupled Neural Network (PCNN) and Nonsubsampled contourlet transform (NSCT), then propose an improved image fusion algorithm based on local entropy, PCNN and NSCT by using the characteristics of PCNN ignition diagram and pixels own characteristics, and combining with the concept of local entropy and matching degree, use local entropy as the resolution of discriminated standard to handle with low frequency coefficients and adopt improved PCNN rules to fuse high frequency coefficients.Related experiments are done in Matlab simulation environment. Improved algorithms separately compare with the traditional Laplace pyramid method, weighted average method based wavelet transform,the local window energy algorithm based wavelet transform and NSCT fusion method. The results show that the two algorithms both have certain image quality’s enhancement and improve the entropy, standard deviation, average gradient and other estimation indexes of image fusion, and they can be more effectively used in medical image or multi-focus image fusion.