Research On Algorithm Of Feature Detection And Classification For Button

The button as part of clothing is being mass-produced every year. There are some inferior-quality products in the production process of the button. The quality inspection of the button is mainly done by manual. The companies which producing button needs slow, inefficient and costly manual inspection to be replaced by intelligent detection equipment. In this thesis, the proposed algorithm based on machine vision system runs on smart camera.The algorithm is design to extract the geometric features and color feature of the button, recognize the blocked or broken buttons and classify the buttons with color.In order to obtain the feature of the button, image segmentation, ROI locating, contour extraction and color space are studied in this thesis. The geometric features such as area,contour diameter, the number of holes,hole diameter, distances between holes and color feature of buttons are extracted accurately by the proposed algorithm. Effective feature vectors are extracted and trained by pattern recognition algorithm to defect button detection and classify button. The feature vectors which definition is the ratio of the variance of button area and the square of the average button area are trained by SVM to recognize the blocked buttons. The broken buttons are recognized by its ratio of variance of the distances and the square of the average distance from the points in the contour to the centet of the button. We use K-means clustering algorithm to train the feature vectors which definition is the average HSV of the button area to classify the buttons with color.The algorithm was transplanted to DSP platform and optimized followed by the verification of the algorithm on PC. Results show that the geometric features and color feature of buttons are extracted accurately by the algorithm at the speed of 9 fps. Both of blocked and broken button detection algorithm achieve the accuracy of 95% at the speed of 20 fps. Buttons are classified accurately with color by the algorithm at the speed of 15 fps.