Sub-pixel Feature Extraction's Key Technology Research And Its Application

Feature extraction is the intermediate section of image processing and the extraction quality affects the result of the location, measurements classified. Recent years, measurement based on high precision is applied in large industrial areas. To improve the detection precision has been a hotspot in machine vision fields, and the sub-pixel application is the important way to achieve high precision measurement and match. The common detection and position precision just can reach to one pixel, with the application such as industry require higher precision, the traditional test methods are difficult to meet actual needs.Improve the hardware's resolution is the most direct way to improve the position precision, however the cost of improving the hardware's resolution is very expensive, therefore, the method using software to analog the pixel between physical pixel is used to make corner, edge detection to sub-pixel in this paper. There are three main work in this article:(1) The sub-pixel edge detection can reach higher precision, this paper combines Canny and Zernike to improve the edge position precision based on the classic pixel edge detection.(2) A fast algorithm for X-corner detection based on special character of X-corners is proposed to solve problem that present corner detection algorithms require too much computation. The algorithm positions original corners used with the edge feature around the X-corners, then according to self-defined symmetry, real corners are detected.(3) In order to improve the precision of corner detection, a sub-pixel corner detector based on outline curvature and local iteration is presented to satisfy the algorithm's stability and real-time. The proposed detector first uses an self-defined edge tracing algorithm and the Canny algorithm to extract the object contour. Secondly, pixel corners are calculated by using the ACSS. Lastly, to center around the pixel corner, sub-pixel corner is checked based on iterated algorithm and bilinear interpolation with principle of vector's orthogonality.Experimental results of feature position and measurement on bin, chess box, clothing shows that: The chess box corner detection algorithm is robust to image noise , works stably under complex environment and achieves excellent effects in online real-time camera calibration; Sub-pixel corners detection method offers a high effective and robust solution to images. At the same time, to bin, chess box and clothing, the precision can up to 0.1 mm, 0.01mm and 0.02mm. It can meet the real-time, high accuracy requirement in practice. At present the detector has been used in the industrial field of embedded machine vision.