Abstract: Circular target localization from a digital image is an important task in computer vision. In this paper, a new algorithm to locate circular objects from degraded gray scale images is proposed. A new type of difference named contour difference is used in this method to detect circular object. The contour difference is calculated directly on the image space, not on the edge map. This procedure needs neither any pre-processing, such as deblurring and denoising, nor edge detection on the input picture. By the new operator, the task of circular target localization becomes a peak searching problem in the space of contour difference data. The experiments show that the proposed method outperforms the circular Hough transform and the recent EDCircles methods on the accuracy in the degraded background, and it is robust to noise and blurs.

Key words: Target localization, contour difference, circular Hough transform, degraded image

摘要： 从数字图像中定位圆形目标是机器视觉领域中的一项重要任务。对常用圆形目标定位方法而言，从降质灰度图像中定位圆形目标面临着准确性较差的问题，针对该问题，提出一种采用轮廓差分概念的圆形目标定位方法。与其他方法相比，该方法不需要对图像进行降噪、反模糊和边缘提取等操作，而是直接从图像平面中计算轮廓差分，通过搜寻轮廓差分的极大值而实现圆形目标的定位。在降质图像上的实验表明，该方法比圆霍夫变换和EDCircles方法对噪声和模糊等降质因素的鲁棒性更好。

关键词: 目标定位, 轮廓差分, 圆霍夫变换, 降质图像