- 期刊
Image Analysis Model based on Sub-pixel Edge Localization
摘要
Image edge analysis is of great significance for image recognition and computer analysis. If the pixel can be subdivided on the original 1-precision pixel, a more accurate localization to the image edge position can be obtained, thus improving the overall measurement accuracy of the image measurement system. In this article, we propose a sub-pixel edge extraction model based on the polynomial interpolation improved Sobel operator to perform image edge analysis on the obtained sub-pixel edge coordinate points. And we have achieved the desired result. we first locate the edge pixel positions with 1 precision using the canny operator. The Sobel operator is applied to the grayscale image to find its corresponding gradient image, and an 8-directional gradient direction template is specified. After obtaining the information of coordinate points and adjacent gradient modulus values as well as gradient directions, the interpolation method is used to obtain subpixel edge points. For images with complex light and other external disturbing factors, we first do pre-process of the image, including binarization, expansion, and erosion of the image. After minimizing the interfering factors, we use the obtained subpixel edge extraction model to extract the contours of the image object edges. Finally, the extracted contour information is analyzed and calculated. consider directly creating a correspondence between pixel coordinates and spatial coordinates. However, the lack of a control parameter prevents the spatial coordinates of the object in the image from corresponding exactly to its pixel coordinates, so a pixel equivalent is calibrated. Firstly, this pixel equivalent is obtained by least-squares fitting and calculation of the calibration circle based on the sub-pixel edge points of the calibration circle in the dot matrix calibration plate. The coordinates of the center of the calibration circle in the calibration plate and the distance between the pixel centers of the two calibration circles are calculated, and the physical dimensions of the centers of the two calibration circles are known, and the ratio of the two is the pixel equivalent. Secondly, the circle centers are constant at a certain object distance and light source intensity. Based on the consideration of hardware factors affecting imaging, we build an error model to achieve a comprehensive calibration of the measurement system using the coordinates of the calibration circle centers. with known image subpixel edge coordinate points, we fit all subpixel coordinate points using the least-squares method to partition the image contour into straight line segments, circular arc segments including circles, and elliptical arc segments including ellipses. The experimental data are first analyzed, and it is known that the image contour is oriented clockwise from the end to the beginning of the data, so it is prescribed to obtain the data from the end to the beginning. The node with large coordinate changes in the data, we consider as the node where the geometric segment changes and the contour is segmented at this node.