A method and system for recognizing text in computer images comprisingdistorted text provides an adaptive iterative process wherein recognitionrules are adapted, added or omitted based on the present state of therecognition process. When the first pass through the recognition andadaptation is completed, the remaining unrecognized words (15) are passedthrough the recognition system 1 using the modified set of recognition rulesstored in (18) and the process is repeated. In most cases the recognitionsystem 1 will identify further reliable recognized words, which iterativelycan be used to improve the recognition rules until the true text comprised inimage (10) is recognized throughout the whole text. The steps of the methodaccording to the present invention are thus repeated until convergence.
展开▼