| 摘要 |
<p>A circuit arrangement for recognising a human face in a sequence of video frames is described. The circuit arrangement contains first means (I) by means of which the video frames are structured in accordance with blocks and by means of which two successive frames are subtracted from one another, and second means (II) for postprocessing of the difference frames. To render the recognition reliable, the areas of a difference frame which could correspond to moving objects are determined by means of comparison means (IIA). Each block of the difference frame which belongs to moving objects is unambiguously mapped to a binary one by means of mapping means (IIB). All other blocks are mapped to a binary zero. This so-called status image of binary ones and zeros is transformed by means of transformation means (IIC) into a one-dimensional attribute function. Using convolution means (IID), it is investigated whether the attribute function determined can be characteristic of a head-shoulder scene. If this is so, marking means (IIE, IIF) are used for marking all points of the status image which are located within a predetermined rectangle into which the assumed face could fall. The analysis is refined to pixel basis by using further means (IIG, IIH, III). <IMAGE></p> |
