Matching And Recognition Based On Industrial CT Image

Industrial CT(Computerized Tomography)is the advanced non-destructive testing technology since it developed from 80's of 20th century . Industrial CT is not affected by the characteristics of the tested objects such as material, shape and structure, and generates intuitionistic images with high resolution, so shows advantages in testing complicated components. Industrial CT technique, acknowledged as the best non-destructive testing and quality evaluation method, plays an important role in aeronautical engineering, space technology and mechanical manufacture.The main defect of industrial CT image can be sorted to key parts'assembly misfit, for example ,whether exist key parts or not ,and object's interior defect such as pore and crack. The study of this dissertation is focus on assembly misfit recognition technology of industrial CT image.This dissertation is devoted to assembly defect detection based on industrial CT. The research work of this dissertation can be classified in the following respects: Firstly, introduces the basic theories of image matching, including concept, mathematic description, process flow, similitude measurement, research contents, existing matching algorithms. Discusses kinds of error factor that can influence the capability of matching, especially analyses and compares the filtering noisy methods.Secondly, as to invariant moment matching, uses a two stage method.In the first stage ,the matching candidates are selected using a computationally low cost feature. Frequency domain calculation is adopted to reduce the computation cost for this stage. In the second stage, invariant moment is performed only on the matching candidates.This algorithm is fast and accurate.Then, studies a square-block searching algorithm for advancing matching speed.The conventional searching algorithm is ergodic, the speed of matching is slow. To improve this situation, model is divided into some sub-model ,calculate eigenvalue of every sub-model. Adopts interlaced scanning for object image, then compares eigenvalue of sub-image with sub-models', choose the best matching position.Finally, due to speciality of industrial CT images, aimed at industrial CT images that including multi-object in per slice, studies a multi-object detect method based arc-radius projection. Through arc-radius projection, limit the images'searching area, namely a part of a sector area. Only computes the matching candidates in this area. Experiment show that this method is available and low cost.