ハフおよびフーリエ変換を用いた拡大・回転・平行移動検出法の部品位置決めへの応用

抄録

This paper discusses on a method of computing the scale of magnification, the angle of rotation, and the quantity of parallel translation that is based on a Hough transform and a Fourier transform. This method applies a Radon transform to images in all directions, differentiates the Radon transformed planes, divides the differentiated planes into positive and negative areas, computes the power spectra of period 2π by applying one-dimensional Fourier transforms to positive and negative differentiated planes, applies a Mellin transform, and gets the scale of magnification and the angle of rotation. Known methods by using 2D Fourier transforms are unable to compute uniquely the angle of rotation from power spectra since they are π-periodic. The proposed method can compute uniquely the angle of rotation from power spectra since they are 2π-periodic. Moreover, to reduce the computational cost, a Radon transform is replaced by a Hough transform by taking account of edges. Experiments were applied to actual images of industrial parts.