A multiple expert algorithm for the binarization of Korean Identity card images

We propose three thresholding methods for binarizing the address region of Korean Identity card images. The first one is a global method which uses a k-means algorithm whose k value is 3, and the second one is a local method which modifies Wellnerpsilas method known as a moving average algorithm. The last one is a multiple expert system which combines three methods selected among a variety of global and local methods including the above two algorithms. A gray pixel in the input image is determined to be black when more than two algorithms out of the three report it as a black. Performance of the proposed algorithms have been evaluated by using a commercial OCR software ARMI 6.0, and we have proved that the results of the proposed algorithms are superior to those of conventional methods.