| 摘要 |
PURPOSE:To obtain a method for dividing a document picture area to discriminate between a linear figure area where a ruled line and a thick line exist and a photograph area by examining the frequency of occurrences of non-linear figure density characteristic patterns. CONSTITUTION:Discrimination of non-linear figure characteristic patterns is closely related to density level division, therefore firstly band division for density level is performed as a preprocessing. Where background density level and black density level fluctuate due to paper quality and printing blur in actual document picture, and so the density band is decided while examining density value occurrence probability of the document picture. Next, according to a scan order, the type of the density pattern of all the mesh by the unit of 3X3 is decided to store the results in a memory. A non-background pattern line appearing between the background patterns is detected as one dimensional arrangement, and whether there is a non-linear figure density characteristic pattern exists in the arrangement is detected. As a result, the method decides as a photograph area if there is said pattern to extract the photograph area from the document picture while discriminating.
|
