Optimal affine image normalization approach for optical character recognition

Optical character recognition (OCR) in images captured from arbitrary angles requires preliminary normalization, i.e. a geometric transformation resulting an image as if it was at angle suitable for OCR. In most cases, surface containing characters can be considered flat, and pinhole model adopted camera. Thus, theory, the normalization should projective. Usually, camera optical axis is approximately perpendicular to document surface, so projective replaced with affine one without significant loss of accuracy. An performed significantly faster than which important OCR on mobile devices. this work, we propose fast approach normalization. It utilizes instead there no The based proposed criterion accuracy: root mean square (RMS) coordinate discrepancies over region interest (ROI). problem optimal according considered. We have established that unconstrained optimization quadratic reduced fractional functions integration ROI. latter solved analytically case where ROI consists rectangles. generalized various cases when transform its special are used: scaling, translation, shearing, their superposition, allowing procedure further accelerated.