Research On Printed Matter Defect Detection Machine Vision System Software Development Based On Component

With the development of computer software and hardware and image processing technology, machine vision technology is widely used in the on-line inspection system in the industry field. While an automatic detection method for printed matter based on machine vision becomes feasible. Machine vision software is an essential software tool in the course of inspecting based on machine vision. At present, foreign software is performing the more effective function, but it usually costs too much, few research staffs or institutes could afford to buy, and because of not knowing the source codes, maintenance becomes difficult. So it is necessary to apply a completely new software developing technique to develop a kind of software system with characteristics of convenience of use, more likely to be combined with some new way of image processing.In this paper, the author presented the machine vision software development method based on component, taking the example of the defect detection system for printed matter and using the platform of Visual C++6.0, designed printed matter defect detection machine vision software based on component. Recults showed that the adoption of software developing technique based on component could reduce developing cost, speed up the whole process of developing, decrease repetitive labor and input, lower software maintenance cost.In the process of developing the algorithm of machine vision software, image matching and Blob analysis algorithm were researched. An improved Harris corner detection method was proposed and testified by a stability evaluation criteria, where the proposed detector was found to perform well in industrial environment. An affine transform model was established to approximate the geometrical changes between two windows of corresponding feature, the parameters of the model was computed by deterministic annealing, in this way, to avoid time-consuming exhaustive searches. The fundamental matrix and homography matrix were estimated robustly with RANSAC algorithm, then epipolar and homography constrains were used to delete the wrong correspondency of initial matching; this paper presented a fast algorithm for blob analysis based on connected components labeling, the algorithm used the method of run-lists and dynamic array, performed just a single scan, needn't to build equivalences table and to unite equivalent labels, in this way, resolved label redundancies in conventional algorithms. In addition, the algorithm also labeled holes in blobs. Experiments showed that with faster speed and good stability, it could correctly detect any blob regions with complicated shapes and random numbers, and computed blobs features.