Fitting of industrial computed tomography images based on edge extracting by neural networks
Changjiang Liu · Computer Integrated Manufacturing Systems · 2010
Through the edge fitting of Industrial Computed Tomography(ICT) images,vector-based curves,and the Computer Aided Design(CAD) drawings of work-pieces could be achieved to realize the reverse design of the inspected work-piece.Based on the image segment with two groups of Cellular Neural Networks(CNN),the edge trace and multidimensional curves fitting were discussed.To fit the cylindrical object,fitting methods of cross circle and the axis line were presented.Through projecting the center coordinates of all floors' circle onto the XZ,YZ plane and fitting respectively by the least square,the computation complexity of the axis line's fitting was reduced.According to fitting parameters,the CAD drawing of the engine slice images' cylindrical object was obtained,and the mean square error was less than 0.3px2.For the irregular objects,the piecewise cubic curve-fitting based on the least square was presented.To fit the engine slice images' target area,the mean square error was all less than 0.6px2.Experimental results and the error analysis validated the fitting methods.And the bitmap's vector required in the reverse design based on ICT was realized.