截图抽奖

README.md

截图抽奖

使用了opencv的python接口

注意使用了非自由的sift/surf（需要编译nonfree的opencv-contrib-python库）

基本思路

• 准备：
• 准备识别的模板（截图抽奖大圆圈）data/tem2x.png
• 准备奖品时间照片放在data/times/*.png
• 计算抽奖目标区域相对于模板的位置和大小
• 计算data/times/*.png的sift/surf，储存在字典
• 计算data/tem2x.png的sift/surf
• 实际计算
• 用skimage.io读取输入文件/url
• 计算输入的sift/surf
• knn比对
• 若比对点少于30视为未找到，放弃
• 否则截取目标区域并计算sift/surf，分别与字典内的数据比对，取最大值为结果

参考

主要代码复制自opencv官网文档 https://docs.opencv.org/3.4.3/dc/dc3/tutorial_py_matcher.html

若干博客内容

维基百科

abdetect.py

import sys,os

from skimage import io
import urllib.request
import numpy as np
import cv2
#from matplotlib import pyplot as plt

dataPath = "C:\\to\\your\datadir"

class AutoBanDectector():
    MIN_MATCH_COUNT = 30
    def __init__(self):
        # init sift and surf
        self.sift = cv2.xfeatures2d.SIFT_create()
        self.surf = cv2.xfeatures2d.SURF_create()
        # build sift result dict
        # todo: pickle it
        self.resDict = {}

        for fn in os.listdir(dataPath+"\\times"):
            pfn = dataPath+"\\times\\"+fn
            nt = fn.split(".")[0]
            img = cv2.imread(pfn,0)
            self.resDict[nt] = self.sift.detectAndCompute(img,None)
        # init flann searcher
        FLANN_INDEX_KDTREE = 0
        index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
        search_params = dict(checks = 50)
        self.flann = cv2.FlannBasedMatcher(index_params, search_params)
        # build surf points for template
        img = cv2.imread(dataPath+"\\tem2x.png",0) # todo: from structed data
        self.kpTemplate, self.desTemplate = self.surf.detectAndCompute(img,None)
    def fuck(self,fn):
        img2d = io.imread(fn,0)
        if None is img2d:
            raise Exception("bad file %s"%fn)
        sca = max(img2d.shape)//1024
        while sca > 1:
            sca = sca>>1
            img2d = cv2.pyrDown(img2d)
        
        kp, des = self.surf.detectAndCompute(img2d,None)
        
        matches = self.flann.knnMatch(self.desTemplate,des,k=2)
        
        # from ocv offical document
        good = []
        for m,n in matches:
            if m.distance < 0.7*n.distance:
                good.append(m)
        print("found",len(good))
        if len(good)>self.MIN_MATCH_COUNT:
            # warp perspective
            src_pts = np.float32([ self.kpTemplate[m.queryIdx].pt for m in good ]).reshape(-1,1,2)
            dst_pts = np.float32([ kp[m.trainIdx].pt for m in good ]).reshape(-1,1,2)
            imgText = cv2.warpPerspective(img2d, cv2.findHomography(dst_pts, src_pts, cv2.RANSAC)[0], (img2d.shape[0],img2d.shape[1]))
            imgText = imgText[0:75, 65:101] # todo: from structed data
            return self.findBestmatch(imgText)
        else:
            # fail
            return None
    def findBestmatch(self,img):
        _,des = self.sift.detectAndCompute(img,None)
        maxMatch = "none"
        maxMatchNum = 0
        for k in self.resDict.keys():
            matches = self.flann.knnMatch(des,self.resDict[k][1],k=2)
            mt = 0
            for m,n in matches:
                if m.distance < 0.7*n.distance:
                    mt+=1
            #print(k,mt)
            if mt>=maxMatchNum:
                maxMatch = k
                maxMatchNum = mt
        #print("bestmatch",maxMatch,maxMatchNum)
        return maxMatch
        
if __name__ == "__main__":
    detector = AutoBanDectector()
    print(detector.fuck(sys.argv[1]))