Character and pattern recognition based on moire images

The paper presents a novel method for recognizing raised or indented characters or patterns on industrial samples by using a combination of moire interferometry technique with optical character recognition (OCR) and pattern recognition. Patterns recognized with this method are of low contrast, and conventional recognition schemes require complex optics and lighting. Raised characters on tires, vin code tags, credit cards, indented characters on metal, wrinkles on skin, and embossment on buttons are some examples. The proposed method uses the moire interferometry technique to obtain a gray scale image of patterns such that their heights are represented in gray scale. This eliminates the need for special optics for each application. 3D images obtained as above, are processed by three sets of algorithms: 1) analytical geometry, 2) pattern recognition, and 3) character recognition. The analytical geometry algorithms consist of constrained and unconstrained fitting methods for scattered data, and transformations between different spaces. The pattern recognition methods consist of feature extraction based on scatter matrices, and classification based on hierarchic classification methods. The OCR algorithm employs gray scale correlation. Extension experiments are conducted to support the method.