Application of Least Square Support Vector Regression in Mechanical Part Linear Edge Sub-pixel Image Detection

In order to improve measurement precision for mechanical part parameters, taking the standard measuring block as the study object in this paper, an application method of least square support vector regression in mechanical part linear edge sub-pixel image detection was put forward. Firstly, digital image of the measuring block was collected by A102FCCD device, and was input into computer by IEEE1394 digital card. Secondly, original gray level image with noise was changed into edge information with single-pixel width after it was processed by reducing noise with median filtering, creating a binary image with threshold method and contour extraction, etc. Thirdly, pixel points on edges in the determined line regions made up the training set of corresponding line, and the least square support vector machine for regression was trained by the training set and regression function of every detected line was obtained, which is the expression of the sub-pixel, and then the width of the measuring block was measured. Linear regression precision acquired by least square support vector regression is higher than linear regression precision acquired by least square linear regression through experiment contrast. Mean square deviations of measuring block width measured by least square support vector regression method is smaller than 0.03 pixels. Finally, the measurement error was analyzed. Theory analysis and experimental results show that the method proposed is characterized by high speed and high precision.