ocr_align.py

def align(truth_lines, ocr_lines):
    nonaligned = []
    aligned = []
    align_idx = 0
    for truth_line in truth_lines:
        best_error = 1.0
        best_align = 0
        for idx, ocr_line in enumerate(ocr_lines[align_idx+1:],
                                        align_idx+1):
            total_error = levenshtein(truth_line, ocr_line)
            error = total_error / len(truth_line)
            if error < best_error:
                best_error = error
                best_align = idx
        if best_error < 0.5:
            num_aligned += 1
            align_idx = best_align
            aligned.append((align_idx, truth_line))
            logger.debug(
                u"Matched `{}` <-> `{}` ({})".format(
                    truth_line, ocr_text[best_align], best_error))
        else:
            nonaligned.append(truth_line)
            if best_align < len(ocr_text):
                logger.debug(
                    u"Could not find match for '{}', closest candidate "
                    u"was '{}' with an error of {}"
                    .format(truth_line, ocr_text[best_align], best_error))
    return aligned, nonaligned