| 摘要 |
A technique is provided for classifying images in a library ( 1 ) for subsequent selective retrieval. Each library image is segmented ( 2 ) into a plurality of coherent regions. Each of the regions is analyzed in respect of a plurality of visual properties, each of which is ascribed a number according to the value of the visual property. The property numbers are then compared ( 4 ) with bands of the available property range and a computer-readable character, such as a letter of the alphabet is ascribed to each of the property values. The characters are then arranged ( 5 ) in a predetermined order to form a region character string. The strings for each region are then indexed ( 7 ) to form an index ( 8 ) of the images by region character string. For image retrieval, a target image ( 10 ) is analyzed in the same way to form region character strings which are then supplied to a text retrieval system ( 12 ) for detecting matches with entries in the index ( 8 ). Library images having matching regions are retrieved from the library ( 1 ).
|
