拼接的拖动验证码距离识别

最近遇到一个图片切割后拼接的滑动验证码，拖动把图片位置还原，要识别他的滑动距离，看起来也不复杂，想着没必要用宰牛刀（机器学习），简单记录一下识别的两种方法：

1. 原理是通过计算像素的欧氏距离来确定上半部分第一张图的长度，像素欧氏距离越小，越趋近同一块位置，说明那个位置是分割点。代码如下:

   class Concat:
   
       def __init__(self, big, rate, draw=True, out="result.png"):
            "rate下半部分图片高度"
           self.draw = draw
           self.out = out
           self.x = 590
           self.y = 360
           self.rate_y = self.y - rate
           self.img = Image.open(BytesIO(big)).convert("RGBA") if isinstance(big, bytes) else Image.open(big).convert("RGBA")
           self.rate_y_x_1 = [self.img.getpixel((i, self.rate_y - 1)) for i in range(0, self.x)]
           self.rate_y_x_2 = [self.img.getpixel((i, self.rate_y)) for i in range(0, self.x)]
   
       def calculate_point_distance(self, x1, x2):
           """ 计算两点间欧氏距离 """
           if len(x1) != len(x2):
               raise Exception("x1 x2 维度不同")
           return math.sqrt(sum([(int(x2[i]) - int(x1[i])) * (int(x2[i]) - int(x1[i])) for i in range(len(x1))]))
   
       def calculate_list_distance(self, list1, list2):
           """ 计算两数组间欧氏距离 """
           distance = 0
           if len(list1) != len(list2):
               raise Exception("list1 list2 长度不同")
           for i in range(len(list1)):
               distance += self.calculate_point_distance(list1[i], list2[i])
           return distance
   
       def discern(self):
           result = []
           for i in range(self.x):
               if i == 0:
                   result.append(self.calculate_list_distance(self.rate_y_x_1, self.rate_y_x_2))
               else:
                   tmp = self.rate_y_x_1.pop(-1)
                   self.rate_y_x_1.insert(0, tmp)
                   result.append(self.calculate_list_distance(self.rate_y_x_1, self.rate_y_x_2))
           result = result.index(min(result))
           if self.draw:
               draw = ImageDraw.Draw(self.img)
               self.draw_pic(draw, [(self.x - result, 0), (self.x - result, 360)], fill='red', width=2, gap=10)
           return self.x - result
   
       def draw_pic(self, draw, pos_list, fill, width, gap):
   
           x_begin, y_begin = pos_list[0]
           x_end, y_end = pos_list[1]
           for _y in range(y_begin, y_end, gap):
               draw.line([(x_begin, _y), (x_begin, _y + gap / 2)], fill=fill, width=width)
           self.img.save(self.out)

   识别效果如下：

2. 先转成灰度图在自适应二值化，然后通过LSD算法找出直线位置坐标，即拼接位置，代码如下：

   import cv2 as cv
   
   
   class Concat:
       def __init__(self, rate, origin):
           self.rate = rate #下半部分图片高度
           self.origin = origin
           self.x = 590    #图片宽高
           self.y = 360
   
       def discren(self):
           origin = cv.imread(self.origin)  # 原图
           src = origin[:self.rate, :, :]  # 切割
           gray = cv.cvtColor(src, cv.COLOR_BGR2GRAY)  # 转为单通道灰度图
           binary = cv.adaptiveThreshold(gray, 255, cv.ADAPTIVE_THRESH_MEAN_C, cv.THRESH_BINARY, 11, 5)  # 自适应阈值二值化
           edges = cv.Canny(binary, 50, 150, apertureSize=3)  # 边缘检测
           lsd = cv.createLineSegmentDetector(0)  # LSD检测
           lines = lsd.detect(edges)
           pos_y = 0  # 直线长度记录
           pos_arr = None  # 直线起始终止点位置数组
           for line in lines[0]:
               x1, y1, x2, y2 = list(map(lambda x: int(round(x)), line[0]))
               if (abs(x2 - x1) < 1) and (abs(y2 - y1) > 20):
                   if abs(y2 - y1) > pos_y:
                       pos_y = abs(y2 - y1)
                       pos_arr = [(x1, y1), (x2, y2)]
           if pos_arr:
               print("图片拼接位置为：", pos_arr[0][0])
               cv.line(origin, (pos_arr[0][0], 0), (pos_arr[1][0], self.x), (0, 255, 0), 2)
               cv.imwrite("new_img.jpg", origin)
               cv.imshow("Original", origin)
               # cv.imshow("Edges", edges)
               cv.waitKey(0)
               cv.destroyAllWindows()
           else:
               raise Exception("没找到位置")
   
   
   if __name__ == '__main__':
       cc = Concat(180, r"bgImage.jpg")
       cc.discren()
   

识别效果如下：