The DR Image Fusion Based On NSCT

The proportion of castings in aviation engines is increasing. Due to the harshenvironment of the engine castings, high quality requirements, and mostly complexstructures, the ratio of max. and min. penetrated thickness which is large when using thetraditional film radiographic testing, the resulting films often carry the information ofoverlapping or interference, and it could easily lead to misjudgment and omission ofjudge of personnel. With the development of computer technology, the DR testingtechnology based on flat panel detector has been widely used in non-destructive testing.In this paper, image fusion technology has been applied to DR images, and the resultingfused images increases by the amount of information compared with the unfused,eliminating the overlapping and interference information between multiple images, thiswill greatly facilitate the assessment of observers, and help promoting the image fusiontechnology in DR testing. This paper entirety takes some aviation engine castings forexample, and it has carried out related work focusing on DR image registration andfusion.Firstly, MSAC algorithm which is better to the robustness of outliers has been usedto do precise matching aiming at the presence of mismatch points after coarse match bynearest neighbor method, and comparison validation is done on simulated images andreal DR images. The results show that: using MSAC algorithm to do precise matchingcan effectively eliminate the mismatch points; registration accuracy based on SURF isin the same order of magnitude with SIFT, but real-time is better than SIFT. Thereforethe registration method based on SURF is more suitable for DR image real-timeregistration.Secondly, the decomposition level of NSCT and the impact of the low-pass filterand fusion rule on fusion image quality is in-depth analyzed aiming at specific DRimages. The results show that: with the increase of decomposition level, the advantageof multi-scale and multi-directional of NSCT has been gradually shown, the fusionquality has gradually improved, however time-consuming has increased with it. Itshould balance the image quality and time-consuming integration; low-pass filter withdifferent bandwidths will influence the information of high frequency components, and DR image decomposition should select the low pass filter of a narrow bandwidth;different fusion rules will affect the quality of fused image, according to the DR imagecharacteristics, taking the low-frequency fusion rules weighted by PCA andhigh-frequency fusion rules of large local gradient can get better fused image.Finally, use the selected NSCT fusion rules and parameters to do fusion on theactual DR images. The results show that: because the fusion method based on NSCTcan effectively avoid ringing, so it is more suitable for DR image fusion respect to thewavelet transform and Contourlet transform; Weighted Fusion Quality Index andEdge-dependent Fusion Index are consistent with subjective assessment results, so thatthey can apply to the fusion evaluation of DR image, and provide a theoretical andexperimental basis for image fusion technology applied in DR testing in the future.