def __init__(self, graph=None, mask=None, navigate_timeout=20):
if graph is not None:
raise ValueError(
"The `graph` parameter of `stbt.Keyboard` constructor is "
"deprecated. See the API documentation for details.")
self.G = nx.DiGraph()
self.G_ = None # navigation without shift transitions that type text
self.modes = set()
self.mask = None
if isinstance(mask, numpy.ndarray):
self.mask = mask
elif mask:
self.mask = load_image(mask)
self.navigate_timeout = navigate_timeout
self.symmetrical_keys = {
"KEY_DOWN": "KEY_UP",
"KEY_UP": "KEY_DOWN",
"KEY_LEFT": "KEY_RIGHT",
"KEY_RIGHT": "KEY_LEFT",
}
self._any_with_region = False
self._any_without_region = False
self._any_with_mode = False
self._any_without_mode = False
def test_that_text_region_is_correct_even_with_regions_larger_than_frame():
frame = load_image("ocr/menu.png")
text, region, _ = list(iterate_menu())[6]
result = stbt.match_text(
text, frame=frame, region=region.extend(right=+12800))
assert result
assert region.contains(result.region)
def test_ocr_doesnt_leak_python_future_newtypes():
f = load_image("ocr/small.png")
result = stbt.ocr(f)
assert type(result).__name__ in ["str", "unicode"]
result = stbt.match_text("Small", f)
assert type(result.text).__name__ in ["str", "unicode"]
def test_that_match_text_gives_tesseract_a_hint():
frame = load_image("ocr/itv-player.png")
if "ITV Player" in stbt.ocr(frame=frame):
raise SkipTest("Tesseract doesn't need a hint")
if "ITV Player" not in stbt.ocr(frame=frame, tesseract_user_words=["ITV"]):
raise SkipTest("Giving tesseract a hint doesn't help")
assert stbt.match_text("ITV Player", frame=frame)
def test_match_text_stringify_result():
frame = load_image("ocr/menu.png")
result = stbt.match_text(u"Onion Bhaji", frame=frame)
assert re.match(
r"TextMatchResult\(time=None, match=True, region=Region\(.*\), "
r"frame=<Image\(filename=u?'ocr/menu.png', "
r"dimensions=1280x720x3\)>, text=u?'Onion Bhaji'\)", str(result))
def test_ocr_text_color(image, color, expected, region):
frame = load_image(image)
mode = stbt.OcrMode.SINGLE_LINE
assert expected not in stbt.ocr(frame, region, mode)
assert expected == stbt.ocr(frame, region, mode, text_color=color)
assert not stbt.match_text(expected, frame, region, mode)
assert stbt.match_text(expected, frame, region, mode, text_color=color)
def test_ocr_on_static_images(image, expected_text, region, mode):
kwargs = {"region": region}
if mode is not None:
kwargs["mode"] = mode
text = stbt.ocr(load_image("ocr/" + image), **kwargs)
assert text == expected_text
# Don't leak python future newtypes
assert type(text).__name__ in ["unicode", "str"]
def test_tesseract_user_patterns(patterns):
# pylint:disable=protected-access
if _tesseract_version() < LooseVersion('3.03'):
raise SkipTest('tesseract is too old')
# Now the real test:
assert u'192.168.10.1' == stbt.ocr(
frame=load_image('ocr/192.168.10.1.png'),
mode=stbt.OcrMode.SINGLE_WORD,
tesseract_user_patterns=patterns)
def test_that_setting_config_options_has_an_effect():
# Unfortunately there are many tesseract config options and they are very
# complicated so it's difficult to write a test that tests that a config
# option is having the correct effect. Due to the difficulty in determining
# "correctness" instead here we test that setting a config option has an
# effect at all. This at least excercises our code which sets config
# options. I'm not happy about this and I hope to be able to replace this
# once we have more experience with these settings in the real world.
if _tesseract_version() >= LooseVersion('3.04'):
hocr_mode_config = {
"tessedit_create_txt": 0,
"tessedit_create_hocr": 1
}
else:
hocr_mode_config = {"tessedit_create_hocr": 1}
assert (stbt.ocr(frame=load_image('ocr/ambig.png'),
tesseract_config=hocr_mode_config) !=
stbt.ocr(frame=load_image('ocr/ambig.png')))
def test_that_match_text_still_returns_if_region_doesnt_intersect_with_frame(
region):
frame = load_image("ocr/menu.png")
result = stbt.match_text("Onion Bhaji", frame=frame, region=region)
assert result.match is False
assert result.region is None
assert result.text == "Onion Bhaji"
# Avoid future.types.newtypes in return values
assert type(result.text).__name__ in ["str", "unicode"]
def test_ocr_text_color_threshold():
f = load_image("ocr/blue-search-white-guide.png")
c = (220, 220, 220)
assert stbt.ocr(f) != "Guide"
# pylint:disable=fixme
# TODO: Find an example where text_color_threshold is necessary. Since
# tesseract 4.0.0 the default text_color_threshold actually works.
# assert stbt.ocr(f, text_color=c) != "Guide"
assert stbt.ocr(f, text_color=c, text_color_threshold=50) == "Guide"
with temporary_config({'ocr.text_color_threshold': '50'}):
assert stbt.ocr(f, text_color=c) == "Guide"
def _test_that_cache_speeds_up_ocr():
frame = load_image('red-black.png')
def ocr():
return stbt.ocr(frame=frame)
# pylint:disable=protected-access
_cache = imgproc_cache._cache
imgproc_cache._cache = None
uncached_result = ocr()
uncached_time = min(timeit.repeat(ocr, repeat=10, number=1))
imgproc_cache._cache = _cache
cached_result = ocr() # prime the cache
cached_time = min(timeit.repeat(ocr, repeat=10, number=1))
print("ocr with cache: %s" % (cached_time, ))
print("ocr without cache: %s" % (uncached_time, ))
assert uncached_time > (cached_time * 10)
assert type(cached_result) == type(uncached_result) # pylint:disable=unidiomatic-typecheck
assert cached_result == uncached_result
r = stbt.Region(x=0, y=32, right=91, bottom=59)
frame2 = load_image("red-black-2.png")
def cached_ocr1():
return stbt.ocr(frame=frame, region=r)
def cached_ocr2():
return stbt.ocr(frame=frame2, region=r)
cached_ocr1() # prime the cache
time1 = timeit.timeit(cached_ocr1, number=1)
time2 = timeit.timeit(cached_ocr2, number=1)
print("ocr with cache (frame 1): %s" % (time1, ))
print("ocr with cache (frame 2): %s" % (time2, ))
assert time2 < (time1 * 10)
assert cached_ocr1() == cached_ocr2()
def test_that_ocr_engine_has_an_effect():
if _tesseract_version() < LooseVersion("4.0"):
raise SkipTest('tesseract is too old')
f = load_image("ocr/ambig.png")
# This is a regression in tesseract 4.0's legacy engine, compared to 3.04:
assert "sillyness" not in stbt.ocr(f, engine=stbt.OcrEngine.TESSERACT)
assert "sillyness" not in stbt.ocr(f)
# ...but the new LSTM engine does read it correctly:
assert "sillyness" in stbt.ocr(f, engine=stbt.OcrEngine.LSTM)
with temporary_config({'ocr.engine': 'LSTM'}):
assert "sillyness" in stbt.ocr(f)
def test_that_ligatures_and_ambiguous_punctuation_are_normalised():
frame = load_image('ocr/ambig.png')
text = stbt.ocr(frame)
for bad, good in [
# tesseract 3.02
("horizonta|", "horizontal"),
# tesseract 4.00 with tessdata 590567f
("siIIyness", "sillyness"),
("Iigatures", "ligatures"),
]:
text = text.replace(bad, good)
assert ligature_text == text
assert stbt.match_text("em-dash,", frame)
assert stbt.match_text(u"em\u2014dash,", frame)
def test_corrections(corrections, expected):
f = load_image('ocr/00.png')
print(corrections)
assert expected == stbt.ocr(frame=f, mode=stbt.OcrMode.SINGLE_WORD,
corrections=corrections)
try:
stbt.set_global_ocr_corrections({'OO': '11'})
if expected == "OO":
expected = "11"
assert expected == stbt.ocr(frame=f, mode=stbt.OcrMode.SINGLE_WORD,
corrections=corrections)
finally:
stbt.set_global_ocr_corrections({})
def test_that_text_location_is_recognised():
frame = load_image("ocr/menu.png")
def test(text, region, upsample):
result = stbt.match_text(text, frame=frame, upsample=upsample)
assert result
assert region.contains(result.region) # pylint:disable=no-member
for text, region, multiline in iterate_menu():
# Don't currently support multi-line comments
if multiline:
continue
yield (test, text, region, True)
yield (test, text, region, False)
def __init__(self, graph, mask=None, navigate_timeout=20):
if isinstance(graph, nx.DiGraph):
self.G = graph
else:
self.G = Keyboard.parse_edgelist(graph)
try:
nx.relabel_nodes(self.G, {"SPACE": " "}, copy=False)
except KeyError: # Node SPACE is not in the graph
pass
_add_weights(self.G)
self.mask = None
if isinstance(mask, numpy.ndarray):
self.mask = mask
elif mask:
self.mask = load_image(mask)
self.navigate_timeout = navigate_timeout
def test_that_ocr_reads_unicode():
text = stbt.ocr(frame=load_image('ocr/unicode.png'), lang='eng+deu')
assert isinstance(text, str)
assert u'£500\nDavid Röthlisberger' == text
def test_that_ocr_can_read_small_text():
text = stbt.ocr(frame=load_image('ocr/small.png'))
assert u'Small anti-aliased text is hard to read\nunless you magnify' == \
text
def test_that_ocr_region_none_isnt_allowed():
with pytest.raises(TypeError):
stbt.ocr(frame=load_image("ocr/small.png"), region=None)
def test_that_ocr_region_none_isnt_allowed(region):
f = load_image("ocr/small.png")
with pytest.raises((TypeError, ValueError)):
stbt.ocr(frame=f, region=region)
with pytest.raises((TypeError, ValueError)):
stbt.match_text("Small", frame=f, region=region)
def test_that_match_text_accepts_unicode():
f = load_image("ocr/unicode.png")
assert stbt.match_text("David", f, lang='eng+deu') # ascii
assert stbt.match_text("Röthlisberger", f, lang='eng+deu') # unicode
assert stbt.match_text("Röthlisberger".encode('utf-8'), f,
lang='eng+deu') # utf-8 bytes
def test_ocr_on_static_images(image, expected_text, region, mode):
kwargs = {"region": region}
if mode is not None:
kwargs["mode"] = mode
text = stbt.ocr(load_image("ocr/" + image), **kwargs)
assert text == expected_text
def test_char_whitelist():
# Without char_whitelist tesseract reads "OO" (the letter oh).
assert u'00' == stbt.ocr(frame=load_image('ocr/00.png'),
mode=stbt.OcrMode.SINGLE_WORD,
char_whitelist="0123456789")
def test_that_default_language_is_configurable():
f = load_image("ocr/unicode.png")
assert not stbt.match_text(u"Röthlisberger", f) # reads Réthlisberger
with temporary_config({"ocr.lang": "deu"}):
assert stbt.match_text(u"Röthlisberger", f)
assert u"Röthlisberger" in stbt.ocr(f)
def test_user_dictionary_with_non_english_language(words):
assert u'192.168.10.1' == stbt.ocr(
frame=load_image('ocr/192.168.10.1.png'),
mode=stbt.OcrMode.SINGLE_WORD,
lang="deu",
tesseract_user_words=words)
def test_ocr_on_text_next_to_image_match():
frame = load_image("action-panel.png")
m = stbt.match("action-panel-blue-button.png", frame)
assert "YOUVIEW MENU" == stbt.ocr(frame,
region=m.region.right_of(width=150))
def test_match_text_case_sensitivity():
frame = load_image("ocr/menu.png", cv2.IMREAD_GRAYSCALE)
assert stbt.match_text("ONION BHAJI", frame)
assert stbt.match_text("ONION BHAJI", frame, case_sensitive=False)
assert not stbt.match_text("ONION BHAJI", frame, case_sensitive=True)
def test_match_text_on_single_channel_image():
frame = load_image("ocr/menu.png", cv2.IMREAD_GRAYSCALE)
assert stbt.match_text("Onion Bhaji", frame)
def test_that_match_text_returns_no_match_for_non_matching_text():
frame = load_image("ocr/menu.png")
assert not stbt.match_text(u"Noodle Soup", frame=frame)
