StarRailCopilot/module/campaign/campaign_ocr.py
LmeSzinc a8a243052f Add: 增加了未通关的章节名的识别
- ocr预处理时会将图片左对齐
2020-04-18 21:20:42 +08:00

165 lines
5.4 KiB
Python

import collections
import numpy as np
from module.base.ocr import Ocr
from module.base.utils import extract_letters, area_offset
from module.logger import logger
from module.template.assets import TEMPLATE_STAGE_CLEAR, TEMPLATE_STAGE_PERCENT, Button
stage_clear_color = tuple(np.mean(np.mean(TEMPLATE_STAGE_CLEAR.image, axis=0), axis=0))
stage_percentage_color = tuple(np.mean(np.mean(TEMPLATE_STAGE_PERCENT.image, axis=0), axis=0))
def ensure_chapter_index(name):
"""
Args:
name (str, int):
Returns:
int
"""
if isinstance(name, int):
return name
else:
if name.isdigit():
return int(name)
elif name in ['a', 'c', 'sp']:
return 1
elif name in ['b', 'd']:
return 2
def separate_name(name):
"""
Args:
name (str): Stage name in lowercase, such as 7-2, d3, sp3.
Returns:
tuple[str]: Campaign_name and stage index in lowercase, Such as ['7', '2'], ['d', '3'], ['sp', '3'].
"""
if '-' in name:
return name.split('-')
elif name.startswith('sp'):
return 'sp', name[-1]
elif name[0] in 'abcdef':
return name[0], name[-1]
class CampaignOcr:
stage = {}
chapter = 0
def extract_campaign_name_image(self, image):
result = TEMPLATE_STAGE_CLEAR.match_multi(image, similarity=0.95)
# np.sort(result.flatten())[-10:]
# array([0.8680386 , 0.8688129 , 0.8693155 , 0.86967576, 0.87012905,
# 0.8705039 , 0.99954903, 0.99983317, 0.99996626, 1. ],
# dtype=float32)
name_offset = (75, 9)
name_size = (60, 16)
name_letter = (255, 255, 255)
name_back = (102, 102, 102)
digits = []
for point in result:
point = point[::-1]
button = tuple(np.append(point, point + TEMPLATE_STAGE_CLEAR.image.shape[:2][::-1]))
point = point + name_offset
name = image.crop(np.append(point, point + name_size))
name = extract_letters(name, letter=name_letter, back=name_back)
stage = self.extract_stage_name(name)
digits.append(Button(area=area_offset(stage, point), color=stage_clear_color, button=button, name='stage'))
result = TEMPLATE_STAGE_PERCENT.match_multi(image, similarity=0.95)
name_offset = (50, 0)
for point in result:
point = point[::-1]
button = tuple(np.append(point, point + TEMPLATE_STAGE_PERCENT.image.shape[:2][::-1]))
point = point + name_offset
name = image.crop(np.append(point, point + name_size))
name = extract_letters(name, letter=name_letter, back=name_back)
stage = self.extract_stage_name(name)
digits.append(
Button(area=area_offset(stage, point), color=stage_percentage_color, button=button, name='stage'))
if len(digits) == 0:
logger.warning('No stage found.')
return digits
@staticmethod
def extract_stage_name(image):
x_skip = 7
interval = 5
x_color = np.convolve(np.mean(image, axis=0), np.ones(interval), 'valid') / interval
x_list = np.where(x_color[x_skip:] > 235)[0]
if x_list is None or len(x_list) == 0:
logger.warning('No interval between digit and text.')
return 0, 0, x_list[0] + 1 + x_skip, image.shape[0]
@staticmethod
def name_separate(image):
"""
Args:
image (np.ndarray): (height, width)
Returns:
list[np.ndarray]:
"""
# Image.fromarray(image.astype('uint8'), mode='L').show()
x_skip = 2
interval = 5
x_color = np.convolve(np.mean(image, axis=0), np.ones(interval), 'valid') / interval
x_list = np.where(x_color[x_skip:] > 235)[0]
if x_list is None or len(x_list) == 0:
logger.warning('No interval between digit and text.')
image = image[:, :x_list[0] + 1 + x_skip]
dash_color_range = (220 - 3, 220 + 3)
dash_height_index = 9
mean = np.mean(image, axis=0)
# print(mean)
x_list = np.where(
(mean > dash_color_range[0])
& (mean < dash_color_range[1])
& (np.argmin(image, axis=0) == dash_height_index)
)[0]
if x_list is None or len(x_list) == 0:
logger.warning('No dash found between digits')
chapter = (0, 0, x_list[0] - 1, image.shape[0])
stage = (x_list[-1] + 1, 0, image.shape[1], image.shape[0])
return chapter, stage
def get_stage_name(self, image):
self.stage = {}
buttons = self.extract_campaign_name_image(image)
ocr = Ocr(buttons, lang='stage')
result = ocr.ocr(image)
if not isinstance(result, list):
result = [result]
result = [res.lower().replace('--', '-') for res in result]
chapter = [separate_name(res)[0] for res in result]
counter = collections.Counter(chapter)
self.chapter = counter.most_common()[0][0]
for name, button in zip(result, buttons):
button.area = button.button
button.name = name
self.stage[name] = button
logger.attr('Chapter', self.chapter)
logger.attr('Stage', ', '.join(self.stage.keys()))
def get_chapter_index(self, image):
"""
A tricky method for ui_ensure_index
"""
self.get_stage_name(image)
return ensure_chapter_index(self.chapter)