| 摘要 |
A method of processing data such as image or audio data comprises the steps of: comparing first and second samples (1, 2) of a space, the first and second samples being related by a set of transformations, to provide a set of pairs of signals, each of the pairs of signals corresponding to substantially the same feature in the space in the first and second samples; for each of the pairs of signals, evaluating (51) residuals corresponding to the differences between the values of the space in each sample; evaluating (52) a first estimate of the set of transformations from the residuals; estimating (53) a tensor field representing orientation of flow in the space, the tensor field covering at least one point in the space; and recalculating (54), from the first estimate and the tensor field, the residuals for each of the pairs of signals to improve the estimate of the set of transformations. The method may be used to detect motion in an image sequence. |
