def get_output_room_type(img: Image) -> Optional[str]:
vw, vh = common.get_vwvh(img)
roi = img.crop(
(52.333 * vh, 1.667 * vh, 88.556 * vh, 8.222 * vh)).array[:, :, :3]
roi_size = roi.shape[0] * roi.shape[1]
low = np.array([30, 182, 245])
high = np.array([40, 192, 255])
mask = cv2.inRange(roi, low, high)
trade_point = np.count_nonzero(mask) / roi_size
low = np.array([245, 210, 0])
high = np.array([255, 220, 10])
mask = cv2.inRange(roi, low, high)
manu_point = np.count_nonzero(mask) / roi_size
low = np.array([195, 230, 65])
high = np.array([205, 240, 75])
mask = cv2.inRange(roi, low, high)
power_point = np.count_nonzero(mask) / roi_size
if trade_point > manu_point + power_point:
return trade
elif manu_point > trade_point + power_point:
return manu
elif power_point > trade_point + manu_point:
return power
else:
return None
def is_building_task_product_button(
img: Image, rect: Tuple[float, float, float, float]) -> bool:
roi = img.crop(rect)
ret, thresh = cv2.threshold(roi.array[:, :, 0], 200, 255,
cv2.THRESH_BINARY_INV)
roi_size = thresh.shape[0] * thresh.shape[1]
return (np.count_nonzero(thresh) / roi_size) > 0.5
def is_need_change_char_reconfirm(
img: Image, rect: Tuple[float, float, float, float]) -> bool:
roi = img.crop(rect)
roi_array = roi.array
roi_size = roi_array.shape[0] * roi_array.shape[1]
tmp_array = (roi_array[:, :, 2] - (roi_array[:, :, 0]) > 100)
return (np.count_nonzero(tmp_array) / roi_size) > 0.5
def is_dormitory_exist(img: Image, rect: Tuple[float, float, float,
float]) -> bool:
roi = img.crop(rect)
ret, thresh = cv2.threshold(roi.array[:, :, 0], 55, 255,
cv2.THRESH_BINARY_INV)
roi_size = thresh.shape[0] * thresh.shape[1]
return (np.count_nonzero(thresh) / roi_size) > 0.5
def get_img_roi(img: Image, rect: Tuple) -> Image:
rect_int = []
for i in rect:
if isinstance(i, float):
rect_int.append(int(i))
else:
rect_int.append(i)
return img.crop(rect)
def is_char_select(img: Image, rect: Tuple[float, float, float,
float]) -> bool:
roi = img.crop(rect)
roi_array = roi.array[:, :, :3]
roi_size = roi_array.shape[0] * roi_array.shape[1]
tmp_array = (roi_array[:, :, 1] >
(roi_array[:, :, 0]) * 1.2) & (roi_array[:, :, 1] >
(roi_array[:, :, 2]) * 1.2)
return (np.count_nonzero(tmp_array) / roi_size) > 0.1
def is_can_start_clue_party(img: Image, rect: Tuple[float, float, float,
float]) -> bool:
roi = img.crop(rect)
roi_array = roi.array[:, :, :3]
low = np.array([255, 255, 255])
high = np.array([255, 255, 255])
mask = cv2.inRange(roi_array, low, high)
roi_size = roi_array.shape[0] * roi_array.shape[1]
return (np.count_nonzero(mask) / roi_size) > 0.5
def get_manu_product_name(img: Image) -> str:
vw, vh = common.get_vwvh(img)
roi = img.crop((21.778 * vh, 85.778 * vh, 43.111 * vh, 90.222 * vh))
eng = ocr.acquire_engine_global_cached('zh-cn')
recognize = eng.recognize(roi,
70,
hints=[ocr.OcrHint.SINGLE_LINE],
char_whitelist=known_product_chars).text.replace(
' ', '')
return recognize
def is_trade_manu_need_change_char(
img: Image, rect: Tuple[float, float, float, float]) -> Optional[bool]:
roi = img.crop(rect).array[:, :, :3]
roi_size = roi.shape[0] * roi.shape[1]
hsv = cv2.cvtColor(roi, cv2.COLOR_RGB2HSV)
if np.count_nonzero(hsv[:, :, 1] > 10) / roi_size < 0.5:
return None
if np.count_nonzero((roi[:, :, 0] > roi[:, :, 1] * 1.2) &
(roi[:, :, 0] > roi[:, :, 2] * 1.2)) / roi_size > 0.6:
return True
return False
def is_clue_get_by_from_friend(
img: Image, rect: Tuple[float, float, float, float]) -> Optional[bool]:
roi = img.crop(rect)
roi_array = roi.array
hsv = cv2.cvtColor(roi_array, cv2.COLOR_RGB2HSV)
hsv_size = hsv.shape[0] * hsv.shape[1]
if (np.count_nonzero(hsv[:, :, 1] > 0) / hsv_size) < 0.1:
return None
roi_array = roi_array[roi_array.shape[0] * 1 // 10:roi_array.shape[0] *
4 // 10, roi_array.shape[1] * 4 // 5:, :3]
low = np.array([124, 0, 33])
high = np.array([126, 2, 35])
mask = cv2.inRange(roi_array, low, high)
roi_size = roi_array.shape[0] * roi_array.shape[1]
return (np.count_nonzero(mask) / roi_size) < 0.05
def is_trade_gold(img: Image) -> bool:
vw, vh = common.get_vwvh(img)
roi = img.crop(
(133.222 * vh, 82.556 * vh, 159.444 * vh, 94.333 * vh)).array[:, :, :3]
roi_size = roi.shape[0] * roi.shape[1]
low = np.array([0, 112, 163])
high = np.array([10, 122, 173])
mask = cv2.inRange(roi, low, high)
gold_point = np.count_nonzero(mask) / roi_size
low = np.array([0, 0, 96])
high = np.array([10, 10, 106])
mask = cv2.inRange(roi, low, high)
not_gold_point = np.count_nonzero(mask) / roi_size
if gold_point > not_gold_point:
return True
else:
return False
def is_clue_on(img: Image, rect: Tuple[float, float, float, float]) -> bool:
hsv = cv2.cvtColor(img.crop(rect).array, cv2.COLOR_RGB2HSV)
roi_size = hsv.shape[0] * hsv.shape[1]
return (np.count_nonzero(hsv[:, :, 1] > 20) / roi_size) > 0.15
def recognize_operator_box(self,
img: Image.Image,
recognize_skill=False,
skill_facility_hint=None) -> OperatorBox:
name_img = img.subview(
(0, 375, img.width, img.height - 2)).convert('L')
name_img = imgops.enhance_contrast(name_img, 90, 220)
name_img = imgops.crop_blackedge2(name_img)
name_img = Image.fromarray(
cv2.copyMakeBorder(255 - name_img.array,
8,
8,
8,
8,
cv2.BORDER_CONSTANT,
value=[255, 255, 255]))
# save image for training ocr
# name_img.save(os.path.join(config.SCREEN_SHOOT_SAVE_PATH, '%d-%04d.png' % (self.tag, self.seq)))
self.seq += 1
# OcrHint.SINGLE_LINE (PSM 7) will ignore some operator names, use raw line for LSTM (PSM 13) here
# the effect of user-words is questionable, it seldom produce improved output (maybe we need LSTM word-dawg instead)
ocrresult = self.ocr.recognize(name_img,
ppi=240,
tessedit_pageseg_mode='13',
user_words_file='operators')
name = ocrresult.text.replace(' ', '')
if name not in operator_set:
comparisions = [(n, textdistance.levenshtein(name, n))
for n in operator_set]
comparisions.sort(key=lambda x: x[1])
self.logger.debug('%s not in operator set, closest match: %s' %
(name, comparisions[0][0]))
if comparisions[0][1] == comparisions[1][1]:
self.logger.warning('multiple fixes availiable for %r',
ocrresult)
name = comparisions[0][0]
mood_img = img.subview(Rect.from_xywh(44, 358, 127,
3)).convert('L').array
mood_img = np.max(mood_img, axis=0)
mask = (mood_img >= 200).astype(np.uint8)
mood = np.count_nonzero(mask) / mask.shape[0] * 24
tagimg = img.subview((35, 209, 155, 262))
on_shift = resources.load_image_cached('riic/on_shift.png', 'RGB')
distracted = resources.load_image_cached('riic/distracted.png', 'RGB')
rest = resources.load_image_cached('riic/rest.png', 'RGB')
tag = None
if imgops.compare_mse(tagimg, on_shift) < 3251:
tag = 'on_shift'
elif imgops.compare_mse(tagimg, distracted) < 3251:
tag = 'distracted'
elif imgops.compare_mse(tagimg, rest) < 3251:
tag = 'rest'
has_room_check = img.subview(Rect.from_xywh(45, 2, 62, 6)).convert('L')
mse = np.mean(np.power(
has_room_check.array.astype(np.float32) - 50, 2))
self.richlogger.logtext(f'has_room_check mse={mse}')
if mse < 200:
room_img = img.subview(Rect.from_xywh(42, 6, 74, 30)).array
room_img = imgops.enhance_contrast(
Image.fromarray(np.max(room_img, axis=2)), 64, 220)
room_img = Image.fromarray(255 - room_img.array)
self.richlogger.logimage(room_img)
room = self.ocr.recognize(
room_img,
ppi=240,
hints=[ocr.OcrHint.SINGLE_LINE],
char_whitelist='0123456789FB').text.replace(' ', '')
else:
room = None
if recognize_skill:
skill1_icon = img.subview(Rect.from_xywh(4, 285, 54, 54))
skill2_icon = img.subview(Rect.from_xywh(67, 285, 54, 54))
skill1, score1 = self.recognize_skill(skill1_icon,
skill_facility_hint)
skill2, score2 = self.recognize_skill(skill2_icon,
skill_facility_hint)
else:
skill1 = None
skill2 = None
skill_icons = []
if skill1 is not None:
skill_icons.append(skill1)
if skill2 is not None:
skill_icons.append(skill2)
self.richlogger.logimage(name_img)
self.richlogger.logtext(repr(ocrresult))
result = OperatorBox(None,
name,
mood,
tag,
room,
skill_icons=skill_icons)
self.richlogger.logtext(repr(result))
return result