Research And Application Of Relative Algorithms In Real-time Automatic Vision Inspection System

Traditionally, visual inspection and quality control are performed by human experts. AltHough humans can do the job better than machines in some cases, they are slower than the machines and get tired quickly. As a new visual inspection technique, the automatic vision inspection (AVI) system has the advantages of high speed, high precision and high reliability, can inspect object without touching and may effectively replace human inspection in many cases. Adopting AVI system in industrial inspection can overcome the weakness of human, and improve automation level and productivity greatly. Now more and more AVI systems have been applied in fields of modern automation manufacture, such as operating monitor, product inspection and quality control.In recent years, with the development of correlative subjects including computer vision, image processing, pattern recognition, AVI technique has been improved greatly. However it still has many difficulties. The works in this dissertation are focused on the research of image processing method in AVI and the AVI system integration. In accord with the requirements and characteristics of AVI system, the research of image processing focuses on real-time requirement, robust requirement and knowledge application, especially the shape properties of object. At last, the AVI system constructed on industry computer is designed for a specific task and how to rationally organize the process of image algorithm is also introduced.The major contributions in this dissertation are as follows:(1) A new kind of connected operator named area reconstruction operator is proposed, which has the fundamental property of interacting with the flat zones in image and uses the information of the object area to filter noise. Area reconstruction operator integrates the merits of area filter and morphological reconstruction. The mask image is simplied on the areas of some discrete flat zones in marker image. Based on the definition of area reconstruction operator, some filters with double parameters are proposed, such as open area reconstruction, area reconstruction H-dome filter, area reconstruction H-dome transform, and area reconstruction H-dome improved transform. In order to meet the requirement of real-time application, a queue-based quick algorithm for area construction operator is presented.(2) Morphological wavelet is a new nonlinear (morphological) multisolution approach. It also may offer some attractive computational advantages while applied in real-time fields. A general and flexible approach for constructing two-dimensionalmulti-channel morphological wavelets applied to image transform is presented. This approach is based on both templates and lifting scheme. The gradient templates are used to extract detail signals in prediction lifting step, and a morphological operator is used as update operator to update the low-pass signals. In order to guarantee nonlinear perfect reconstruction, a sufficient condition for template group selection is also discussed.(3) The Circle Hough Transform (CHT) is one of the most popular image algorithms for detecting circle object. However, its fast implementation has many drawbacks while applied in digital images. After analysis of the fast CHT, a new method is presented to offer more robust and accurate identification of circles. The geometric property that the midnormal of symmetry points on the circumference of a circle passes through the center of circle is used to filter non-circular feature points and determine the searching direction of radius. The proposed technique is efficient in computation and robust to noise.(4) In order to efficiently separate the overlapping particles in the image, three new methods are proposed for different cases. In binary image, a method including twice distance-transforms is applied to get more accurate boundary. In grayscale image, after foreground is extracted from image, particles are segmented by intensity features. In the last method, after two kind markers are fusing, overlapping particles are separated by both intensity...