Study On A Fusion Model Driven By Discrepant Features Of Dual-color MWIR Images

Dual-color is the primary means to enhance the performance of Mid-wave infrared(MWIR) imaging, which is widely used in the target detection process of military and civilfield. Moreover, image fusion can combine the advantages of dual-band imaging, and is thekey technology to obtain high-quality images. Therefore, exploring image fusion method withadaptive capacity of Dual-color MWIR has important practical significance and applicationvalue.In order to overcome the difficulty that traditional fusion methods can not dynamicallychange with image discrepant information, a fusion model driven by discrepant features ismade in this paper. Formation mechanism of discrepant features on Dual-color MWIR imagesis analyzed firstly, and then, two methods called “priority-reduction, posteriority-show” and“priority-show, posteriority-reduction” are used to reduce the dimensions of discrepantfeatures. Afterward, fusion demand of dual-band images is defined by researching distributionrules on compositive discrepant features after dimension reduction. Besides, through adoptingsliding window method as processing measure and possibility theory as basis, mappingrelationship between discrepant features and fusion rules is built, which is processed on thebasis of sliding window with fixed size. As a result, fusion model driven by discrepantfeatures is established. To solve the incorrectness problem of fusion model bringing with fixedwindow, a new method of sliding window with adaptive size is presented, and number ofimage edge is regarded as decision condition to determine the size of sliding window. Fromthe above, fusion model will be improved.The experiment shows that the presented fusion model driven by discrepant featuresachieves significant results for the fusion of dual-color MWIR images, which is better thantraditional fusion method both in subjective perception and objective evaluation. Furthermore,the effects of fusion model with adaptive sliding window are more clear and natural.