image.py 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318
  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. def __init__(self, roi_front: tuple, roi_back: tuple, method: str, threshold: float, file: str) -> None:
  13. """
  14. 初始化
  15. :param roi_front: 前置roi
  16. :param roi_back: 后置roi 用于匹配的区域
  17. :param method: 匹配方法 "Template matching"
  18. :param threshold: 阈值 0.8
  19. :param file: 相对路径, 带后缀
  20. """
  21. self._match_init = False # 这个是给后面的 等待图片稳定
  22. self._image = None # 这个是匹配的目标
  23. self._kp = None #
  24. self._des = None
  25. self.method = method
  26. self.roi_front: list = list(roi_front)
  27. self.roi_back = roi_back
  28. self.threshold = threshold
  29. self.file = file
  30. @cached_property
  31. def name(self) -> str:
  32. """
  33. :return:
  34. """
  35. return Path(self.file).stem.upper()
  36. def __str__(self):
  37. return self.name
  38. __repr__ = __str__
  39. def __eq__(self, other):
  40. return str(self) == str(other)
  41. def __hash__(self):
  42. return hash(self.name)
  43. def __bool__(self):
  44. return True
  45. def load_image(self) -> None:
  46. """
  47. 加载图片
  48. :return:
  49. """
  50. if self._image is not None:
  51. return
  52. img = cv2.imdecode(fromfile(self.file, dtype=uint8), -1)
  53. img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
  54. self._image = img
  55. height, width, channels = self._image.shape
  56. if height != self.roi_front[3] or width != self.roi_front[2]:
  57. self.roi_front[2] = width
  58. self.roi_front[3] = height
  59. logger.info(f"roi_front size changed to {width}x{height}")
  60. def load_kp_des(self) -> None:
  61. if self._kp is not None and self._des is not None:
  62. return
  63. self._kp, self._des = self.sift.detectAndCompute(self.image, None)
  64. @property
  65. def image(self):
  66. """
  67. 获取图片
  68. :return:
  69. """
  70. if self._image is None:
  71. self.load_image()
  72. return self._image
  73. @cached_property
  74. def is_template_match(self) -> bool:
  75. """
  76. 是否是模板匹配
  77. :return:
  78. """
  79. return self.method == "Template matching"
  80. @cached_property
  81. def is_sift_flann(self) -> bool:
  82. return self.method == "Sift Flann"
  83. @cached_property
  84. def sift(self):
  85. return cv2.SIFT_create()
  86. @cached_property
  87. def kp(self):
  88. if self._kp is None:
  89. self.load_kp_des()
  90. return self._kp
  91. @cached_property
  92. def des(self):
  93. if self._des is None:
  94. self.load_kp_des()
  95. return self._des
  96. def corp(self, image: np.array, roi: list = None) -> np.array:
  97. """
  98. 截取图片
  99. :param image:
  100. :param roi
  101. :return:
  102. """
  103. if roi is None:
  104. x, y, w, h = self.roi_back
  105. else:
  106. x, y, w, h = roi
  107. x, y, w, h = int(x), int(y), int(w), int(h)
  108. return image[y:y + h, x:x + w]
  109. def match(self, image: np.array, threshold: float = None) -> bool:
  110. """
  111. :param threshold:
  112. :param image:
  113. :return:
  114. """
  115. if threshold is None:
  116. threshold = self.threshold
  117. if not self.is_template_match:
  118. return self.sift_match(image)
  119. # raise Exception(f"unknown method {self.method}")
  120. source = self.corp(image)
  121. mat = self.image
  122. res = cv2.matchTemplate(source, mat, cv2.TM_CCOEFF_NORMED)
  123. 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
  124. # logger.attr(self.name, max_val)
  125. if max_val > threshold:
  126. self.roi_front[0] = max_loc[0] + self.roi_back[0]
  127. self.roi_front[1] = max_loc[1] + self.roi_back[1]
  128. return True
  129. else:
  130. return False
  131. def match_all(self, image: np.array, threshold: float = None, roi: list = None) -> list[tuple]:
  132. """
  133. 区别于match,这个是返回所有的匹配结果
  134. :param roi:
  135. :param image:
  136. :param threshold:
  137. :return:
  138. """
  139. if roi is not None:
  140. self.roi_back = roi
  141. if threshold is None:
  142. threshold = self.threshold
  143. if not self.is_template_match:
  144. raise Exception(f"unknown method {self.method}")
  145. source = self.corp(image)
  146. mat = self.image
  147. results = cv2.matchTemplate(source, mat, cv2.TM_CCOEFF_NORMED)
  148. locations = np.where(results >= threshold)
  149. matches = []
  150. for pt in zip(*locations[::-1]): # (x, y) coordinates
  151. score = results[pt[1], pt[0]]
  152. # 得分, x, y, w, h
  153. x = self.roi_back[0] + pt[0]
  154. y = self.roi_back[1] + pt[1]
  155. matches.append((score, x, y, mat.shape[1], mat.shape[0]))
  156. return matches
  157. def coord(self) -> tuple:
  158. """
  159. 获取roi_front的随机的点击的坐标
  160. :return:
  161. """
  162. x, y, w, h = self.roi_front
  163. return x + np.random.randint(0, w), y + np.random.randint(0, h)
  164. def coord_more(self) -> tuple:
  165. """
  166. 获取roi_back的随机的点击的坐标
  167. :return:
  168. """
  169. x, y, w, h = self.roi_back
  170. return x + np.random.randint(0, w), y + np.random.randint(0, h)
  171. def front_center(self) -> tuple:
  172. """
  173. 获取roi_front的中心坐标
  174. :return:
  175. """
  176. x, y, w, h = self.roi_front
  177. return int(x + w//2), int(y + h//2)
  178. def test_match(self, image: np.array):
  179. if self.is_template_match:
  180. return self.match(image)
  181. if self.is_sift_flann:
  182. return self.sift_match(image, show=True)
  183. def sift_match(self, image: np.array, show=False) -> bool:
  184. """
  185. 特征匹配,同样会修改 roi_front
  186. :param image: 是游戏的截图,就是转通道后的截图
  187. :param show: 测试用的
  188. :return:
  189. """
  190. source = self.corp(image)
  191. kp, des = self.sift.detectAndCompute(source, None)
  192. # 参数1:index_params
  193. # 对于SIFT和SURF,可以传入参数index_params=dict(algorithm=FLANN_INDEX_KDTREE, trees=5)。
  194. # 对于ORB,可以传入参数index_params=dict(algorithm=FLANN_INDEX_LSH, table_number=6, key_size=12)。
  195. index_params = dict(algorithm=1, trees=5)
  196. # 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。
  197. search_params = dict(checks=50)
  198. # 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
  199. flann = cv2.FlannBasedMatcher(index_params, search_params)
  200. # 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
  201. # 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
  202. # 第一个参数是小图的des, 第二个参数是大图的des
  203. matches = flann.knnMatch(self.des, des, k=2)
  204. good = []
  205. result = True
  206. for i, (m, n) in enumerate(matches):
  207. # 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.
  208. if m.distance < 0.6 * n.distance:
  209. good.append(m)
  210. if len(good) >= 10:
  211. src_pts = float32([self.kp[m.queryIdx].pt for m in good]).reshape(-1, 1, 2)
  212. dst_pts = float32([kp[m.trainIdx].pt for m in good]).reshape(-1, 1, 2)
  213. # 计算透视变换矩阵m, 要求点的数量>=4
  214. m, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
  215. # 创建一个包含模板图像四个角坐标的数组
  216. w, h = self.roi_front[2], self.roi_front[3]
  217. pts = float32([[0, 0], [0, h - 1], [w - 1, h - 1], [w - 1, 0]]).reshape(-1, 1, 2)
  218. if m is None:
  219. result = False
  220. else:
  221. dst = int32(cv2.perspectiveTransform(pts, m))
  222. self.roi_front[0] = dst[0, 0, 0] + self.roi_back[0]
  223. self.roi_front[1] = dst[0, 0, 1] + self.roi_back[1]
  224. if show:
  225. cv2.polylines(source, [dst], isClosed=True, color=(0, 0, 255), thickness=2)
  226. if not is_approx_rectangle(np.array([pos[0] for pos in dst])):
  227. result = False
  228. else:
  229. result = False
  230. # https://blog.csdn.net/cungudafa/article/details/105399278
  231. # https://blog.csdn.net/qq_45832961/article/details/122776322
  232. if show:
  233. # 准备一个空的掩膜来绘制好的匹配
  234. mask_matches = [[0, 0] for i in range(len(matches))]
  235. # 向掩膜中添加数据
  236. for i, (m, n) in enumerate(matches):
  237. if m.distance < 0.6 * n.distance: # 理论上0.7最好
  238. mask_matches[i] = [1, 0]
  239. img_matches = cv2.drawMatchesKnn(self.image, self.kp, source, kp, matches, None,
  240. matchColor=(0, 255, 0), singlePointColor=(255, 0, 0),
  241. matchesMask=mask_matches, flags=0)
  242. cv2.imshow(f'Sift Flann: {self.name}', img_matches)
  243. cv2.waitKey(0)
  244. cv2.destroyAllWindows()
  245. return result
  246. def match_mean_color(self, image, color: tuple, bias=10) -> bool:
  247. """
  248. :param image:
  249. :param color: rgb
  250. :param bias:
  251. :return:
  252. """
  253. image = self.corp(image)
  254. average_color = cv2.mean(image)
  255. # logger.info(f'{self.name} average_color: {average_color}')
  256. for i in range(3):
  257. if abs(average_color[i] - color[i]) > bias:
  258. return False
  259. return True
  260. if __name__ == "__main__":
  261. from dev_tools.assets_test import detect_image
  262. IMAGE_FILE = './log/test/QQ截图20240223151924.png'
  263. from tasks.Restart.assets import RestartAssets
  264. jade = RestartAssets.I_HARVEST_JADE
  265. jade.method = 'Sift Flann'
  266. sign = RestartAssets.I_HARVEST_SIGN
  267. sign.method = 'Sift Flann'
  268. print(jade.roi_front)
  269. detect_image(IMAGE_FILE, jade)
  270. detect_image(IMAGE_FILE, sign)
  271. print(jade.roi_front)