image.py 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328
  1. # This Python file uses the following encoding: utf-8
  2. # @author runhey
  3. # github https://github.com/runhey
  4. import cv2
  5. import numpy as np
  6. from numpy import float32, int32, uint8, fromfile
  7. from pathlib import Path
  8. from module.base.decorator import cached_property
  9. from module.logger import logger
  10. from module.base.utils import is_approx_rectangle
  11. class RuleImage:
  12. debug_mode: bool = False
  13. def __init__(self, roi_front: tuple, roi_back: tuple, method: str, threshold: float, file: str) -> None:
  14. """
  15. 初始化
  16. :param roi_front: 前置roi
  17. :param roi_back: 后置roi 用于匹配的区域
  18. :param method: 匹配方法 "Template matching"
  19. :param threshold: 阈值 0.8
  20. :param file: 相对路径, 带后缀
  21. """
  22. self._match_init = False # 这个是给后面的 等待图片稳定
  23. self._image = None # 这个是匹配的目标
  24. self._kp = None #
  25. self._des = None
  26. self.method = method
  27. self.roi_front: list = list(roi_front)
  28. self.roi_back = roi_back
  29. self.threshold = threshold
  30. self.file = file
  31. @cached_property
  32. def name(self) -> str:
  33. """
  34. :return:
  35. """
  36. return Path(self.file).stem.upper()
  37. def __str__(self):
  38. return self.name
  39. __repr__ = __str__
  40. def __eq__(self, other):
  41. return str(self) == str(other)
  42. def __hash__(self):
  43. return hash(self.name)
  44. def __bool__(self):
  45. return True
  46. def load_image(self) -> None:
  47. """
  48. 加载图片
  49. :return:
  50. """
  51. if self._image is not None:
  52. return
  53. img = cv2.imdecode(fromfile(self.file, dtype=uint8), -1)
  54. img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
  55. self._image = img
  56. height, width, channels = self._image.shape
  57. if height != self.roi_front[3] or width != self.roi_front[2]:
  58. self.roi_front[2] = width
  59. self.roi_front[3] = height
  60. logger.debug(f"{self.name} roi_front size changed to {width}x{height}")
  61. def load_kp_des(self) -> None:
  62. if self._kp is not None and self._des is not None:
  63. return
  64. self._kp, self._des = self.sift.detectAndCompute(self.image, None)
  65. @property
  66. def image(self):
  67. """
  68. 获取图片
  69. :return:
  70. """
  71. if self._image is None:
  72. self.load_image()
  73. return self._image
  74. @cached_property
  75. def is_template_match(self) -> bool:
  76. """
  77. 是否是模板匹配
  78. :return:
  79. """
  80. return self.method == "Template matching"
  81. @cached_property
  82. def is_sift_flann(self) -> bool:
  83. return self.method == "Sift Flann"
  84. @cached_property
  85. def sift(self):
  86. return cv2.SIFT_create()
  87. @cached_property
  88. def kp(self):
  89. if self._kp is None:
  90. self.load_kp_des()
  91. return self._kp
  92. @cached_property
  93. def des(self):
  94. if self._des is None:
  95. self.load_kp_des()
  96. return self._des
  97. def corp(self, image: np.array, roi: list = None) -> np.array:
  98. """
  99. 截取图片
  100. :param image:
  101. :param roi
  102. :return:
  103. """
  104. if roi is None:
  105. x, y, w, h = self.roi_back
  106. else:
  107. x, y, w, h = roi
  108. x, y, w, h = int(x), int(y), int(w), int(h)
  109. return image[y:y + h, x:x + w]
  110. def match(self, image: np.array, threshold: float = None) -> bool:
  111. """
  112. :param threshold:
  113. :param image:
  114. :return:
  115. """
  116. if threshold is None:
  117. threshold = self.threshold
  118. if not self.is_template_match:
  119. return self.sift_match(image)
  120. # raise Exception(f"unknown method {self.method}")
  121. source = self.corp(image)
  122. mat = self.image
  123. if mat is None or mat.shape[0] == 0 or mat.shape[1] == 0:
  124. logger.error(f"Template image is invalid: {mat.shape}") #检测模板尺寸,不合法则不进行匹配,避免两次截图画面完全相同造成模板不合法
  125. return True # 如果模板图像无效,直接返回 True
  126. res = cv2.matchTemplate(source, mat, cv2.TM_CCOEFF_NORMED)
  127. min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(res) # 最小匹配度,最大匹配度,最小匹配度的坐标,最大匹配度的坐标
  128. if self.debug_mode:
  129. logger.attr(self.name, f'matching score {max_val:.5f}')
  130. if max_val > threshold:
  131. self.roi_front[0] = max_loc[0] + self.roi_back[0]
  132. self.roi_front[1] = max_loc[1] + self.roi_back[1]
  133. return True
  134. else:
  135. return False
  136. def match_all(self, image: np.array, threshold: float = None, roi: list = None) -> list[tuple]:
  137. """
  138. 区别于match,这个是返回所有的匹配结果
  139. :param roi:
  140. :param image:
  141. :param threshold:
  142. :return:
  143. """
  144. if roi is not None:
  145. self.roi_back = roi
  146. if threshold is None:
  147. threshold = self.threshold
  148. if not self.is_template_match:
  149. raise Exception(f"unknown method {self.method}")
  150. source = self.corp(image)
  151. mat = self.image
  152. results = cv2.matchTemplate(source, mat, cv2.TM_CCOEFF_NORMED)
  153. locations = np.where(results >= threshold)
  154. matches = []
  155. for pt in zip(*locations[::-1]): # (x, y) coordinates
  156. score = results[pt[1], pt[0]]
  157. # 得分, x, y, w, h
  158. x = self.roi_back[0] + pt[0]
  159. y = self.roi_back[1] + pt[1]
  160. matches.append((score, x, y, mat.shape[1], mat.shape[0]))
  161. return matches
  162. def coord(self) -> tuple:
  163. """
  164. 获取roi_front的随机的点击的坐标
  165. :return:
  166. """
  167. x, y, w, h = self.roi_front
  168. return x + np.random.randint(0, w), y + np.random.randint(0, h)
  169. def coord_more(self) -> tuple:
  170. """
  171. 获取roi_back的随机的点击的坐标
  172. :return:
  173. """
  174. x, y, w, h = self.roi_back
  175. return x + np.random.randint(0, w), y + np.random.randint(0, h)
  176. def front_center(self) -> tuple:
  177. """
  178. 获取roi_front的中心坐标
  179. :return:
  180. """
  181. x, y, w, h = self.roi_front
  182. return int(x + w//2), int(y + h//2)
  183. def test_match(self, image: np.array):
  184. self.debug_mode = True
  185. if self.is_template_match:
  186. return self.match(image)
  187. if self.is_sift_flann:
  188. return self.sift_match(image, show=True)
  189. def sift_match(self, image: np.array, show=False) -> bool:
  190. """
  191. 特征匹配,同样会修改 roi_front
  192. :param image: 是游戏的截图,就是转通道后的截图
  193. :param show: 测试用的
  194. :return:
  195. """
  196. source = self.corp(image)
  197. kp, des = self.sift.detectAndCompute(source, None)
  198. # 参数1:index_params
  199. # 对于SIFT和SURF,可以传入参数index_params=dict(algorithm=FLANN_INDEX_KDTREE, trees=5)。
  200. # 对于ORB,可以传入参数index_params=dict(algorithm=FLANN_INDEX_LSH, table_number=6, key_size=12)。
  201. index_params = dict(algorithm=1, trees=5)
  202. # 参数2:search_params 指定递归遍历的次数,值越高结果越准确,但是消耗的时间也越多。
  203. search_params = dict(checks=50)
  204. # 根据设置的参数创建特征匹配器 指定匹配的算法和kd树的层数,指定返回的个数
  205. flann = cv2.FlannBasedMatcher(index_params, search_params)
  206. # 利用创建好的特征匹配器利用k近邻算法来用模板的特征描述符去匹配图像的特征描述符,k指的是返回前k个最匹配的特征区域
  207. # 返回的是最匹配的两个特征点的信息,返回的类型是一个列表,列表元素的类型是Dmatch数据类型,具体是什么我也不知道
  208. # 第一个参数是小图的des, 第二个参数是大图的des
  209. matches = flann.knnMatch(self.des, des, k=2)
  210. good = []
  211. result = True
  212. for i, (m, n) in enumerate(matches):
  213. # 设定阈值, 距离小于对方的距离的0.7倍我们认为是好的匹配点.
  214. if m.distance < 0.6 * n.distance:
  215. good.append(m)
  216. if len(good) >= 10:
  217. src_pts = float32([self.kp[m.queryIdx].pt for m in good]).reshape(-1, 1, 2)
  218. dst_pts = float32([kp[m.trainIdx].pt for m in good]).reshape(-1, 1, 2)
  219. # 计算透视变换矩阵m, 要求点的数量>=4
  220. m, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
  221. # 创建一个包含模板图像四个角坐标的数组
  222. w, h = self.roi_front[2], self.roi_front[3]
  223. pts = float32([[0, 0], [0, h - 1], [w - 1, h - 1], [w - 1, 0]]).reshape(-1, 1, 2)
  224. if m is None:
  225. result = False
  226. else:
  227. dst = int32(cv2.perspectiveTransform(pts, m))
  228. self.roi_front[0] = dst[0, 0, 0] + self.roi_back[0]
  229. self.roi_front[1] = dst[0, 0, 1] + self.roi_back[1]
  230. if show:
  231. cv2.polylines(source, [dst], isClosed=True, color=(0, 0, 255), thickness=2)
  232. if not is_approx_rectangle(np.array([pos[0] for pos in dst])):
  233. result = False
  234. else:
  235. result = False
  236. # https://blog.csdn.net/cungudafa/article/details/105399278
  237. # https://blog.csdn.net/qq_45832961/article/details/122776322
  238. if show:
  239. # 准备一个空的掩膜来绘制好的匹配
  240. mask_matches = [[0, 0] for i in range(len(matches))]
  241. # 向掩膜中添加数据
  242. for i, (m, n) in enumerate(matches):
  243. if m.distance < 0.6 * n.distance: # 理论上0.7最好
  244. mask_matches[i] = [1, 0]
  245. img_matches = cv2.drawMatchesKnn(self.image, self.kp, source, kp, matches, None,
  246. matchColor=(0, 255, 0), singlePointColor=(255, 0, 0),
  247. matchesMask=mask_matches, flags=0)
  248. cv2.imshow(f'Sift Flann: {self.name}', img_matches)
  249. cv2.waitKey(0)
  250. cv2.destroyAllWindows()
  251. return result
  252. def match_mean_color(self, image, color: tuple, bias=10) -> bool:
  253. """
  254. :param image:
  255. :param color: rgb
  256. :param bias:
  257. :return:
  258. """
  259. image = self.corp(image)
  260. average_color = cv2.mean(image)
  261. # logger.info(f'{self.name} average_color: {average_color}')
  262. for i in range(3):
  263. if abs(average_color[i] - color[i]) > bias:
  264. return False
  265. return True
  266. if __name__ == "__main__":
  267. from dev_tools.assets_test import detect_image
  268. IMAGE_FILE = './log/test/QQ截图20240223151924.png'
  269. from tasks.Restart.assets import RestartAssets
  270. jade = RestartAssets.I_HARVEST_JADE
  271. jade.method = 'Sift Flann'
  272. sign = RestartAssets.I_HARVEST_SIGN
  273. sign.method = 'Sift Flann'
  274. print(jade.roi_front)
  275. detect_image(IMAGE_FILE, jade)
  276. detect_image(IMAGE_FILE, sign)
  277. print(jade.roi_front)