| 摘要 |
A technique is provided for segmenting a pixellated image (7) into foreground and background regions. The background and foreground regions of a preceding image are analysed (3) so as to provide noise and signal histograms. A first threshold is derived (24) as the lowest level such that a predetermined proportion of the background pixels have lower noise levels. A second threshold is derived (25) as the highest level such that a predetermined proportion of the foreground pixels have higher signal levels. A difference image is formed (5) as a difference between each pixel of the image (7) and the corresponding pixel of a non-occluded background image (1b). Each of the difference image pixels is allocated (11) as a background pixel if the value of the difference image pixel is less than the first threshold and the pixel is surrounded by a majority of neighbouring difference image pixels which are allocated to the background region. Each of the remaining pixels is then allocated (12) to the foreground region if its value is greater than the second threshold and it has a majority of neighbouring pixels which are allocated to the foreground region. The remaining pixels are then allocated (13) as candidate foreground or background pixels by comparison with a third threshold (26) which lies between the first and second thresholds. Each candidate pixel is then allocated as a background pixel if a majority of its neighbouring pixels are already allocated to the background region or are candidate background pixels. Otherwise, the pixel is allocated to the foreground <IMAGE> |
