| 摘要 |
The invention relates to a system (100) for extracting an object (Ob) from a source image, said object being delineated by a contour (C), the system (100) comprising a gradient unit (110) for computing the source image gradient field, based on the source image, a smoothing unit (120) for smoothing the source image gradient field, and an integration unit (130) for calculating an object image by integrating the smoothed source image gradient field, thereby extracting the object (Ob) from the source image. At each point of the source image, the smoothing is defined by a 2-dimensional convolution kernel which is a product of a first 1-dimensional convolution kernel in the first direction substantially parallel to the contour (C), and a second 1-dimensional convolution kernel in the second direction substantially normal to the contour (C). The first 1-dimensional convolution kernel defines smoothing within each region separated by the contour, while the second 1-dimensional convolution kernel defines smoothing across the contour separating two regions, independently of the orientation of the object and the contour curvature. |
| 主权项 |
1. A system for extracting an object from a source image, said object being delineated by a contour, the system comprising:
a gradient unit that computes the source image gradient field, based on the source image, a smoothing unit that smoothes the source image gradient field, a coordinate unit that:
parameterizes the contour, andtransforms Cartesian coordinates of a point (x, y) in the source image into contour-based coordinates (s, n), wherein s(x, y) is a value of a parameter of the point (x, y)on the parameterized contour corresponding to an intersection of a normal to the contour extending through the point (x, y), wherein n(x, y) is a signed distance between the point (x, y) in the source image and the contour, and wherein the signed distance assumes one sign for points on one side of the contour and an opposite sign for points on the other side of the contour and wherein a first 1-dimensional convolution kernel is dependent on the s coordinate and a second 1-dimensional convolution kernel is dependent on the n coordinate, and an integration unit that calculates an object image by integrating the smoothed source image gradient field, thereby extracting the object from the source image, wherein at each point of the source image, the smoothing is defined by a 2-dimensional convolution kernel which is a product of the first 1-dimensional convolution kernel in a first direction substantially parallel to the contour, and of the second 1-dimensional convolution kernel in a second direction substantially normal to the contour. |
