Spaces:
Running on Zero
Running on Zero
File size: 6,318 Bytes
a39757f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 | import pytest
import numpy as np
import cv2
from app.domain.schemas.core import TextBlock, BoundingBox, OCRResult
from app.services.ocr.evaluator import detect_script, evaluate_detection_quality, select_or_merge_passes
from app.api.dependencies.ocr import get_ocr_engine
# ---------------------------------------------------------
# 1. Rich Raw Detection Schema
# ---------------------------------------------------------
def test_rich_raw_detections():
bbox = BoundingBox(
points=[(10.0, 20.0), (110.0, 20.0), (110.0, 50.0), (10.0, 50.0)],
normalized_points=[(0.01, 0.02), (0.11, 0.02), (0.11, 0.05), (0.01, 0.05)],
width=100.0,
height=30.0,
area=3000.0
)
block = TextBlock(
box=bbox,
text="29501011234567",
confidence=0.98,
x=10.0,
y=20.0,
width=100.0,
height=30.0,
page=1,
detection_order=1,
script="numeric"
)
assert block.x == 10.0
assert block.y == 20.0
assert block.width == 100.0
assert block.height == 30.0
assert block.page == 1
assert block.detection_order == 1
assert block.script == "numeric"
# ---------------------------------------------------------
# 2. Script Detection Per Block
# ---------------------------------------------------------
def test_script_detection_per_block():
assert detect_script("جمهورية مصر العربية") == "ar"
assert detect_script("Commercial Invoice") == "latin"
assert detect_script("123456789") == "numeric"
assert detect_script("Cairo القاهرة 2024") == "mixed"
assert detect_script("###$$$") == "symbol"
# ---------------------------------------------------------
# 3. Multi-Signal Quality Evaluation
# ---------------------------------------------------------
def test_multi_signal_quality_evaluation():
# A. High quality result
bbox = BoundingBox(points=[(0, 0), (10, 0), (10, 10), (0, 10)], normalized_points=[(0,0),(0.1,0),(0.1,0.1),(0,0.1)], width=10, height=10, area=100)
good_blocks = [
TextBlock(box=bbox, text="Total Amount: 15000 EGP", confidence=0.95),
TextBlock(box=bbox, text="Invoice Number: INV-9812", confidence=0.94),
TextBlock(box=bbox, text="Date: 2024-05-15", confidence=0.96)
]
good_res = OCRResult(blocks=good_blocks, total_time_ms=100, average_confidence=0.95, image_width=500, image_height=500)
eval_good = evaluate_detection_quality(good_res)
assert eval_good["composite_score"] > 0.65
assert eval_good["has_structured_data"] is True
assert eval_good["coherence_ratio"] > 0.80
# B. Noisy gibberish result
bad_blocks = [
TextBlock(box=bbox, text="~!!@#$$%", confidence=0.25),
TextBlock(box=bbox, text="^^^^||||", confidence=0.20)
]
bad_res = OCRResult(blocks=bad_blocks, total_time_ms=100, average_confidence=0.22, image_width=500, image_height=500)
eval_bad = evaluate_detection_quality(bad_res)
assert eval_bad["composite_score"] < 0.40
assert eval_bad["has_structured_data"] is False
# ---------------------------------------------------------
# 4. Multi-Pass Selection & Complementary Merging
# ---------------------------------------------------------
def test_multi_pass_selection_and_merging():
bbox1 = BoundingBox(points=[(10, 10), (100, 10), (100, 30), (10, 30)], normalized_points=[(0,0),(0,0),(0,0),(0,0)], width=90, height=20, area=1800)
bbox2 = BoundingBox(points=[(10, 50), (100, 50), (100, 70), (10, 70)], normalized_points=[(0,0),(0,0),(0,0),(0,0)], width=90, height=20, area=1800)
bbox_footer = BoundingBox(points=[(10, 400), (200, 400), (200, 430), (10, 430)], normalized_points=[(0,0),(0,0),(0,0),(0,0)], width=190, height=30, area=5700)
# Pass A found header (bbox1, bbox2)
pass_a = OCRResult(
blocks=[
TextBlock(box=bbox1, text="Company Header", confidence=0.90),
TextBlock(box=bbox2, text="Invoice 1234", confidence=0.88)
],
total_time_ms=80,
average_confidence=0.89,
image_width=500,
image_height=500
)
# Pass B found header (overlapping) AND footer (bbox_footer)
pass_b = OCRResult(
blocks=[
TextBlock(box=bbox1, text="Company Header", confidence=0.89),
TextBlock(box=bbox_footer, text="Tax Registration: 987654321", confidence=0.92)
],
total_time_ms=80,
average_confidence=0.90,
image_width=500,
image_height=500
)
merged, strategy = select_or_merge_passes(pass_a, pass_b)
assert strategy == "merged"
# Merged should have 3 blocks: header + invoice + footer!
assert len(merged.blocks) == 3
assert any("Tax Registration" in b.text for b in merged.blocks)
# ---------------------------------------------------------
# 5. EasyOCR Engine Production Inference
# ---------------------------------------------------------
def test_easyocr_engine_rich_inference():
# Synthetic image with Arabic and English
img = np.ones((400, 800, 3), dtype=np.uint8) * 255
cv2.putText(img, "Tax Invoice 2024", (50, 150), cv2.FONT_HERSHEY_SIMPLEX, 1.2, (0, 0, 0), 2)
cv2.putText(img, "Total: 1500 EGP", (50, 250), cv2.FONT_HERSHEY_SIMPLEX, 1.2, (0, 0, 0), 2)
engine = get_ocr_engine()
res = engine.process_image(img)
assert res is not None
assert len(res.blocks) >= 2
# Check rich attributes
for block in res.blocks:
assert block.width > 0
assert block.height > 0
assert block.page == 1
assert block.detection_order >= 1
assert block.script in ["latin", "numeric", "mixed", "ar", "symbol"]
def test_easyocr_engine_preserves_raw_text_without_post_correction(monkeypatch):
from app.infrastructure.ocr.easyocr_engine import EasyOCREngine
bbox = [[10, 10], [150, 10], [150, 40], [10, 40]]
engine = object.__new__(EasyOCREngine)
engine._initialized = True
class Reader:
def readtext(self, *_args, **_kwargs):
return [(bbox, "محمل", 0.91)]
engine.reader = Reader()
result = engine.process_image(np.ones((100, 200, 3), dtype=np.uint8) * 255)
assert result.blocks[0].text == "محمل"
assert result.blocks[0].raw_text == "محمل"
assert result.blocks[0].normalized_text == "محمل"
|