"""OCR normalization, snapping to the known part set, and occurrence matching.""" import pytest from spareparts import matcher from spareparts.ocr import Detection KNOWN = frozenset({"30642", "74012", "74506", "74639", "74401", "73290", "30638"}) def detection(text, conf=0.99, box=(100, 500, 200, 540)): x0, y0, x1, y1 = box return Detection(text, conf, x0, y0, x1, y1, [[x0, y0], [x1, y0], [x1, y1], [x0, y1]]) @pytest.mark.parametrize( "raw,expected", [ ("*7.4012", "74012"), # spurious dot inside an asterisked callout ("-74506", "74506"), # leader line absorbed into the OCR box ("*30638", "30638"), # footnote marker ("30642", "30642"), ("Seite 18", "18"), ("D / PU-S-12", "12"), ], ) def test_normalize_strips_everything_but_digits(raw, expected): assert matcher.normalize(raw) == expected def test_snap_exact_and_fuzzy(): assert matcher.snap("30642", KNOWN) == ("30642", "exact") # A dropped leading digit, seen on real pages ('4639' for 74639). assert matcher.snap("4639", KNOWN) == ("74639", "fuzzy1") assert matcher.snap("4401", KNOWN) == ("74401", "fuzzy1") def test_snap_refuses_to_invent_part_numbers(): assert matcher.snap("", KNOWN) is None assert matcher.snap("12345", KNOWN) is None # two edits from anything known assert matcher.snap("061101", KNOWN) is None # a drawing revision code assert matcher.snap("1207", KNOWN) is None # 'Seite 12' + section '07' def test_snap_drops_ambiguous_matches(): """Two candidates one edit away is not a match, it is a coin toss.""" ambiguous = frozenset({"74011", "74012"}) assert matcher.snap("7401", ambiguous) is None def test_header_and_footer_strips_are_furniture(): height = 2352 assert matcher.is_furniture(detection("PU-ELA", box=(28, 24, 140, 65)), height) assert matcher.is_furniture(detection("Seite 18", box=(1543, 2290, 1669, 2323)), height) assert not matcher.is_furniture(detection("30642", box=(425, 1644, 520, 1677)), height) def test_labels_for_page_accepts_callouts_and_reports_rejects(): detections = [ detection("30642", box=(425, 1644, 520, 1677)), detection("*7.4012", box=(1243, 190, 1351, 231)), detection("Seite 18", box=(1543, 2290, 1669, 2323)), detection("061101", box=(30, 1000, 140, 1040)), ] labels, rejected = matcher.labels_for_page( detections, KNOWN, frozenset({"30642"}), image_height=2352 ) assert [lab.part_no for lab in labels] == ["30642", "74012"] assert labels[1].has_asterisk is True assert labels[1].raw_text == "*7.4012" # the raw read is preserved assert labels[0].in_page_table is True assert labels[1].in_page_table is False # drawn but absent from this table assert {why for _, _, why in rejected} == {"page furniture", "no part match"} def test_fuzzy_match_needs_higher_confidence(): low = [detection("4639", conf=0.6, box=(400, 900, 500, 940))] labels, rejected = matcher.labels_for_page(low, KNOWN, frozenset(), 2352) assert labels == [] assert rejected[0][2] == "fuzzy below threshold" high = [detection("4639", conf=0.95, box=(400, 900, 500, 940))] labels, _ = matcher.labels_for_page(high, KNOWN, frozenset(), 2352) assert labels[0].part_no == "74639" assert labels[0].match_kind == "fuzzy1" # A fuzzy snap is a weaker claim than a clean read. assert labels[0].confidence < labels[0].ocr_confidence def test_assign_instances_numbers_repeated_callouts(): labels, _ = matcher.labels_for_page( [ detection("74506", box=(200, 100, 300, 140)), detection("74506", box=(200, 900, 300, 940)), detection("30642", box=(700, 500, 800, 540)), ], KNOWN, frozenset(), 2352, ) numbered = matcher.assign_instances(labels) by_part = {} for instance, label in numbered: by_part.setdefault(label.part_no, []).append(instance) assert by_part["74506"] == [0, 1] assert by_part["30642"] == [0]