4cb7c975f5
Add test for 1×1 image (sub-tile-size) resilience and narrow preprocess_page fallback from except Exception to (cv2.error, ValueError, MemoryError) so programming errors propagate instead of being silently swallowed. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
83 lines
3.2 KiB
Python
83 lines
3.2 KiB
Python
"""Tests for the image preprocessing pipeline."""
|
||
|
||
import numpy as np
|
||
import pytest
|
||
from PIL import Image
|
||
from unittest.mock import patch
|
||
|
||
|
||
def _make_yellowed_image(width=100, height=100):
|
||
"""Dark, faded yellowed page: L values in a narrow low range with spatial noise.
|
||
|
||
Very dark (R≈30, G≈20, B≈10) → L_cv ≈ 80-100 in OpenCV uint8 LAB space.
|
||
The per-pixel noise gives each CLAHE tile a non-trivial histogram to equalize,
|
||
which stretches the narrow dark range toward [0-255] and reliably increases mean L.
|
||
"""
|
||
rng = np.random.default_rng(42)
|
||
arr = np.zeros((height, width, 3), dtype=np.uint8)
|
||
arr[:, :, 0] = np.clip(30 + rng.integers(-8, 9, (height, width)), 0, 255)
|
||
arr[:, :, 1] = np.clip(20 + rng.integers(-5, 6, (height, width)), 0, 255)
|
||
arr[:, :, 2] = np.clip(10 + rng.integers(-3, 4, (height, width)), 0, 255)
|
||
return Image.fromarray(arr.astype(np.uint8), mode="RGB")
|
||
|
||
|
||
class TestPreprocessPage:
|
||
def test_output_has_same_dimensions_as_input(self):
|
||
from preprocessing import preprocess_page
|
||
|
||
img = Image.new("RGB", (150, 200))
|
||
result = preprocess_page(img)
|
||
|
||
assert result.size == img.size
|
||
|
||
def test_l_channel_mean_increases_on_yellowed_image(self):
|
||
"""CLAHE equalizes the dark narrow-range histogram toward [0-255], raising mean L."""
|
||
from preprocessing import preprocess_page
|
||
import cv2
|
||
|
||
img = _make_yellowed_image()
|
||
|
||
arr_before = np.array(img)
|
||
lab_before = cv2.cvtColor(arr_before, cv2.COLOR_RGB2LAB)
|
||
l_mean_before = float(lab_before[:, :, 0].mean())
|
||
|
||
result = preprocess_page(img)
|
||
|
||
# Output is grayscale (mode "L"); its values ARE the CLAHE-enhanced L channel
|
||
l_mean_after = float(np.array(result).mean())
|
||
|
||
assert l_mean_after > l_mean_before
|
||
|
||
def test_does_not_crash_on_sub_tile_size_image(self):
|
||
"""A 1×1 image is smaller than the CLAHE tile (8×8) in both axes.
|
||
preprocess_page must not raise — it either succeeds or falls back silently."""
|
||
from preprocessing import preprocess_page
|
||
|
||
img = Image.new("RGB", (1, 1), color=(128, 100, 80))
|
||
result = preprocess_page(img)
|
||
|
||
assert isinstance(result, Image.Image)
|
||
|
||
def test_falls_back_to_pixel_identical_original_on_cv2_error(self):
|
||
"""When cv2 raises a known error, preprocess_page returns the unmodified original image."""
|
||
from preprocessing import preprocess_page
|
||
|
||
img = Image.new("RGB", (80, 60), color=(123, 45, 67))
|
||
original_pixels = list(img.getdata())
|
||
|
||
with patch("preprocessing.cv2.cvtColor", side_effect=ValueError("bad input")):
|
||
result = preprocess_page(img)
|
||
|
||
result_pixels = list(result.getdata())
|
||
assert result_pixels == original_pixels
|
||
|
||
def test_unexpected_exception_propagates(self):
|
||
"""A RuntimeError (programming error) must propagate — not be swallowed by the cv2 fallback."""
|
||
from preprocessing import preprocess_page
|
||
|
||
img = Image.new("RGB", (80, 60))
|
||
|
||
with patch("preprocessing.cv2.cvtColor", side_effect=RuntimeError("unexpected")):
|
||
with pytest.raises(RuntimeError, match="unexpected"):
|
||
preprocess_page(img)
|