python +opencv 多尺度缩放与旋转的模板匹配

#python opencv 多尺度，平移，缩放，旋转等模板匹配法
##多尺度缩放与旋转的均为模板图
import cv2
import numpy as np
import pandas as pd
import time


# 图片旋转函数-保持图像不被裁剪且去除黑边
def ImageRotate(img, angle,borderValue=255):   # img:输入图片；newIm：输出图片；angle：旋转角度(°)
    height, width = img.shape[:2]  # 输入(H,W,C)，取 H，W 的值
    center = (width // 2, height // 2)  # 绕图片中心进行旋转
    M = cv2.getRotationMatrix2D(center, -angle, 1.0)
    cos = np.abs(M[0, 0])
    sin = np.abs(M[0, 1])
    nW = int((height * sin) + (width * cos))
    nH = int((height * cos) + (width * sin))
    M[0, 2] += (nW / 2) - center[0]
    M[1, 2] += (nH / 2) - center[1]
    image_rotation = cv2.warpAffine(img, M, (nW, nH),borderValue=borderValue)
    return image_rotation

#金字塔下采样---弃用（没进行深入理解）
def ImagePyrDown(image,NumLevels):
    for i in range(NumLevels):
        image = cv2.pyrDown(image)       #pyrDown下采样
    return image


# 旋转匹配函数（输入参数分别为模板图像、待匹配图像）
def RatationMatch(modelpicture, searchpicture):
    searchtmp, modeltmp = searchpicture.copy(),modelpicture.copy()

    # 使用matchTemplate对原始灰度图像和图像模板进行匹配
    mask = np.zeros_like(modeltmp, np.uint8)
    pts = np.where(modeltmp == 0)
    mask[pts[0], pts[1]] = 255
    print('searchtmp:',searchtmp.shape,modeltmp.shape)
    res = cv2.matchTemplate(searchtmp, modeltmp, cv2.TM_CCOEFF_NORMED,mask)
    min_val, max_val, min_indx, max_indx = cv2.minMaxLoc(res)
    location = max_indx
    temp = max_val
    angle = 0  # 当前旋转角度记录为0

    tic = time.time()
    # 以步长为5进行第一次粗循环匹配
    for i in range(-90, 181, 5):
        newIm = ImageRotate(modeltmp, i)
        mask = np.zeros_like(newIm, np.uint8)
        pts = np.where(newIm == 0)
        mask[pts[0], pts[1]] = 255
        #cv2.imshow('--mask', mask)
        #cv2.waitKey(0)
        #cv2.destroyAllWindows()
        #mask[newIm[:]==0]=255
        res = cv2.matchTemplate(searchtmp, newIm, cv2.TM_SQDIFF_NORMED,mask)
        min_val, max_val, min_indx, max_indx = cv2.minMaxLoc(res)
        #print(min_val, max_val, min_indx, max_indx)
        if max_val > temp:
            location = max_indx
            temp = max_val
            angle = i
            #print('angle:',angle)
            #draw_result(searchtmp, newIm, (location[0], location[1]))
            #modelpicture00 = ImageRotate(modelpicture, angle)
            #cv2.imshow('--modelpicture00', modelpicture00)
            #cv2.waitKey(0)
            #cv2.destroyAllWindows()
    toc = time.time()
    #ss = ImageRotate(modelpicture, angle)
    #draw_result(searchpicture, ss, (location[0],location[1]))

    print('第一次粗循环匹配所花时间为：' + str(1000 * (toc - tic)) + 'ms'+'------angle:'+str(angle))

    tic = time.time()
    # 在当前最优匹配角度周围10的区间以1为步长循环进行循环匹配计算
    for j in range(angle - 5, angle + 6):
        newIm = ImageRotate(modeltmp, j)
        #print('jj:',j)
        res = cv2.matchTemplate(searchtmp, newIm, cv2.TM_SQDIFF_NORMED)
        min_val, max_val, min_indx, max_indx = cv2.minMaxLoc(res)
        if max_val > temp:
            location = max_indx
            temp = max_val
            angle = j
            #print('angel2:',angle)
            #draw_result(searchtmp, newIm, (location[0], location[1]))
    toc = time.time()
    print('在当前最优匹配角度周围10的区间以1为步长循环进行循环匹配所花时间为：' + str(1000 * (toc - tic)) + 'ms'+'-----angle:'+str(angle))

    tic = time.time()
    # 在当前最优匹配角度周围2的区间以0.1为步长进行循环匹配计算
    k_angle = angle - 1
    for k in range(0, 19):
        k_angle = k_angle + 0.1
        #print('k_angle:',k_angle)
        newIm = ImageRotate(modeltmp, k_angle)
        res = cv2.matchTemplate(searchtmp, newIm, cv2.TM_SQDIFF_NORMED)
        min_val, max_val, min_indx, max_indx = cv2.minMaxLoc(res)
        if max_val > temp:
            location = max_indx
            temp = max_val
            angle = k_angle
            #draw_result(searchtmp, newIm, (location[0], location[1]))
    toc = time.time()
    print('在当前最优匹配角度周围2的区间以0.1为步长进行循环匹配所花时间为：' + str(1000 * (toc - tic)) + 'ms'+"---angle:"+str(angle))
    location_x = location[0]
    location_y = location[1]

    #angle = -angle
    match_point = {'angle': angle,"temp":temp, 'point': (location_x, location_y)}
    #print(match_point)
    return match_point

def scale_resize(img,scale):
    # 要缩小图像，建议选择：cv2.INTER_AREA；如果要放大图像，cv2.INTER_CUBIC效果更好但是速度慢
    if scale <= 1:
        method = cv2.INTER_AREA
    else:
        method = cv2.INTER_CUBIC
    newh = int(img.shape[1] * scale)
    neww = int(img.shape[0] * scale)
    new_dim = [neww, newh]
    resizedimg = cv2.resize(img, new_dim, method)
    return resizedimg

def run_main(templeimg, img):
    imgW,imgH = img.shape[0],img.shape[1]

    #初试：使用matchTemplate对原始灰度图像和图像模板进行匹配
    mask = np.zeros_like(templeimg, np.uint8)
    pts = np.where(templeimg == 0)
    mask[pts[0], pts[1]] = 255
    res = cv2.matchTemplate(img, templeimg, cv2.TM_CCOEFF_NORMED,mask)
    min_val, max_val, min_indx, max_indx = cv2.minMaxLoc(res)
    location_best = max_indx
    temp_best = max_val
    angle_best = 0  # 当前旋转角度记录为0
    best_scale_img = img.copy()
    #print(location_best)

    for scale in np.linspace(3.0, 0, 100)[::-1]:
        #要缩小图像，建议选择：cv2.INTER_AREA；如果要放大图像，cv2.INTER_CUBIC效果更好但是速度慢
        if scale<=1:
            method = cv2.INTER_AREA
        else:
            method = cv2.INTER_CUBIC
        newh = int(templeimg.shape[1]*scale)
        neww = int(templeimg.shape[0] * scale )
        new_dim = [neww,newh]

        # 业务要求：目标文件长度至少为img的宽的二分之一
        if max(neww,newh) <(min(img.shape[0],img.shape[1])/2):
            continue
        if (neww>imgW) or(newh>imgH) :
            continue

        resizedtempleimg = cv2.resize(templeimg, new_dim, method)
        resizedtempleimgcopy = resizedtempleimg.copy()
        # ------旋转匹配 match_point = {'angle': angle,"temp":temp, 'point': (location_x, location_y)}
        match_points = RatationMatch(resizedtempleimgcopy, img)
        temp = match_points['temp']
        angle = match_points['angle']
        point = match_points['point']
        if temp_best 
  
#传统的方法，还是需要大量的参数阈值控制，太鸡肋了，决定，只在数据量很少的冷启动项目阶段采用传统方式。