摘要 |
This method relates to image processing. Each hypercontour is modelled locally by a portion of hypersurface, for example a plane whose parameters Ri can take s sets of predetermined values. For each point of the image, the method consists in: - calculating convolution products L * Gi and L' * G'i, for i = 1 to s, where L and L' are square matrices made up respectively of luminance values of the current image and of the next image; where Gi and G'i are matrices allowing the detection of a hypercontour which can be modelled by a hypersurface passing through the current point and having Ri as parameter values; - calculating values VRi of a generalised likelihood ratio: VRi = ¦ L' * G'i + L * Gi ¦ for i = 1 to s; - determining the value VRS, among values VRi, which is greatest and comparing it with a fixed value lambda ; - concluding that there is no hypercontour passing through the current point if VRS </= lambda , or concluding that there is a hypercontour, which can be modelled by a portion of hypersurface having parameter values corresponding to VRS if VRS > lambda . Application to estimating the motion of a contour element, to determining a three-dimensional contour, etc. <IMAGE> |