Automatic product profile extraction method in visual detection

Abstract

The invention relates to an automatic product profile extraction method in visual detection. The method comprises the following steps: step one, preprocessing an extracted image; step two, extractingan edge in the image; step three, searching for a regional contour in the edge image; to be specific, extracting all enclosed profiles in the edge image, filling all the enclosed profiles to obtain corresponding regions, calculating areas of all regions, finding out an area with the largest area and thus using the corresponding enclosed profile as a product profile; step four, carrying out morphological opening operation on the contour region and removing a vent hole or a bur on the profile; and step five, carrying out product profile restoration. According to the invention, with utilization of a multi-scale edge extraction operator and combination of methods like image gray correction ad morphological calculation methods, a good effect of product region extraction is realized in most of application scenes, so that the need of automatic product region extraction in visual detection of the printing industry is met; automatic extraction of the product region is realized; the step of manual product region drawing is removed; and the application range is wide.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for automatically extracting product outlines in visual inspection. Background technique [0002] Printed matter detection is generally divided into two steps in the specific implementation: the first is the modeling part, that is, the standard printed matter image is obtained through the image collection of qualified products, and the template for detection is established with the standard image, which divides the standard image into different parts. The detection area, and the corresponding detection algorithm is specified for each different detection area. Secondly, it is the process of inspection execution. During this process, the camera collects the images to be inspected in real time, inspects the products to be inspected according to the previously established template, and performs final binning processing according to the inspection results. [0003] ...

AI Technical Summary

Problems solved by technology

However, the manual modeling process is cumbersome and requires the modeling operator to have certain professional knowledge

[0004] The closest prior art has following two kinds: 1, adopt edge detection operators such as sobel, canny to detect the edge of image, divide image by edge; 2, adopt classifiers such as neural network or support vector machine, classify image The product area and the non-product area are trained, and after the training results converge, the trained model is used to classify the product area and the non-product area of ​​the image, so as to realize the automatic extraction of the product contour area; but the above two technologies have the following disadvantages: 1 , Direct use of canny and sobel operators can only extract the edge of the image, and cannot area the actual product outline in the image and other disturbances in the image, such as the outline of texture and vents, etc.

In addition, this method cannot automatically adjust the threshold of edge detection, and cannot automatically adapt to different light sources and automatic extraction of product areas under images collected by different cameras; Extract product area

Training on a variety of different images may cause the classifier to fail to converge

The classifier method is poor for classifying product and non-product regions in black and white images

Images