The Study Of Objective Evaluation Metrics Of Color Image Fusion

At present, in the field of color image fusion, it is impossible to obtain standardfused image as reference. Therefore, it is critical to have the ability to quantify thevisual perception and other useful information contained for subsequent processingof the fused color image. In this study, based on the research of color image fusionalgorithms, within the framework of quaternion theory, we mainly focus on theconstruction of objective evaluation indexes for both color image fusion and thefusion of color and infared images, which were verified by experiments.This study starts by reviewing existing color image fusion technology andintroduces some of the most commonly adopted color image fusion algorithms.Then,a novel multifocus color image fusion algorithm based on the quaternion wavelettransform (QWT) is proposed. Each source color image is represented in thequaternion wavelet transform domain. The different image fusion rules are applied tothe coarse and fine level QWT coefficients to construct a final multiresolutionrepresentation. The fused color image is reconstructed by inverse quaternion wavelettransform.Next, an advanced objective quality assessment method for color image fusionis proposed, which is a no-reference index based on quaternion convolution theory.This method models the color images under quaternion structure and fully utilizes the color information contained in the images as a whole. Detailed peripheralinformation is extracted through a sophisticated quaternion convolution algorithmusing quaternion-valued edge detection template, the original images and the fusedimage. By measuring the clarity and useful information extracted above, theobjective assessment index for the fused color image is established.As for the fusion of color and infrared images, under the research of image localstructural similarity comparison based on the quaternion matrix singular valuedecomposition, the edge information transferring to the fused image is measured asto balance the sensitivity of the assessment methods, and then, the objectiveassessment index for the fusion of color and infrared images is constructed based onthe above procedure.In summary, by fully considering the color information contained in colorimages under quaternion structure, the proposed algorithms are consistent with thehuman visual system. Results from experiments show that the proposed image fusionmethod based on quaternion wavelet transform could overcome the image blurproblem and outperforms the classical methods. Meanwhile, the proposed objectiveassessment methods perfom better compared to some of the commonly adoptedevaluation methods by providing good consistency to HVS and exhibiting goodstability, which is achieved by incorporating both the color information from theimages and detailed information sensitive to the human visual system.