Python+OpenCV实现图像识别替换功能详解

OpenCV-Python是一个Python库，旨在解决计算机视觉问题。

OpenCV是一个开源的计算机视觉库，1999年由英特尔的Gary Bradski启动。Bradski在访学过程中注意到，在很多优秀大学的实验室中，都有非常完备的内部公开的计算机视觉接口。这些接口从一届学生传到另一届学生，对于刚入门的新人来说，使用这些接口比重复造轮子方便多了。这些接口可以让他们在之前的基础上更有效地开展工作。OpenCV正是基于为计算机视觉提供通用接口这一目标而被策划的。

安装opencv

pip3 install -i https:https:///pypi.doubanio.com/simple/ opencv-python

思路：

1、首先区分三张图片：

base图片代表初始化图片；

template图片代表需要在大图中匹配的图片；

white图片为需要替换的图片。

2、然后template图片逐像素缩小匹配，设定阈值，匹配度到达阈值的图片，判定为在初始图片中；否则忽略掉。

3、匹配到最大阈值的地方，返回该区域的位置（x,y）

4、然后用white图片resize到相应的大小，填补到目标区域。

match函数：

"""检查模板图片中是否包含目标图片"""
def make_cv2(photo1, photo2):
    global x, y, w, h, num_1,flag
    starttime = datetime.datetime.now()
    #读取base图片
    img_rgb = cv2.imread(f'{photo1}')
    #读取template图片
    template = cv2.imread(f'{photo2}')
    h, w = template.shape[:-1]
    print('初始宽高', h, w)
    res = cv2.matchTemplate(img_rgb, template, cv2.TM_CCOEFF_NORMED)
    print('初始最大相似度', res.max())
    threshold = res.max()
    """，相似度小于0.2的，不予考虑；相似度在[0.2-0.75]之间的，逐渐缩小图片"""
    print(threshold)
    while threshold >= 0.1 and threshold <= 0.83:
        if w >= 20 and h >= 20:
            w = w - 1
            h = h - 1
            template = cv2.resize(
                template, (w, h), interpolation=cv2.INTER_CUBIC)
            res = cv2.matchTemplate(img_rgb, template, cv2.TM_CCOEFF_NORMED)
            threshold = res.max()
            print('宽度:', w, '高度:', h, '相似度:', threshold)
        else:
            break
    """达到0.75覆盖之前的图片"""
    if threshold > 0.8:
        loc = np.where(res >= threshold)
        x = int(loc[1])
        y = int(loc[0])
        print('覆盖图片左上角坐标:', x, y)
        for pt in zip(*loc[::-1]):
            cv2.rectangle(
                img_rgb, pt, (pt[0] + w, pt[1] + h), (255, 144, 51), 1)
        num_1 += 1
        endtime = datetime.datetime.now()
        print("耗时:", endtime - starttime)
        overlay_transparent(x, y, photo1, photo3)
    else:
        flag = False

replace函数：

"""将目标图片镶嵌到指定坐标位置"""
def overlay_transparent(x, y, photo1, photo3):
    #覆盖图片的时候上下移动的像素空间
    y += 4
    global w, h, num_2
    background = cv2.imread(f'{photo1}')
    overlay = cv2.imread(f'{photo3}')
    """缩放图片大小"""
    overlay = cv2.resize(overlay, (w, h), interpolation=cv2.INTER_CUBIC)
    background_width = background.shape[1]
    background_height = background.shape[0]
    if x >= background_width or y >= background_height:
        return background
    h, w = overlay.shape[0], overlay.shape[1]
    if x + w > background_width:
        w = background_width - x
        overlay = overlay[:, :w]
    if y + h > background_height:
        h = background_height - y
        overlay = overlay[:h]
    if overlay.shape[2] < 4:
        overlay = np.concatenate([overlay, np.ones((overlay.shape[0], overlay.shape[1], 1), dtype=overlay.dtype) * 255],axis=2,)
    overlay_image = overlay[..., :3]
    mask = overlay[..., 3:] / 255.0
    background[y:y + h,x:x + w] = (1.0 - mask) * background[y:y + h,x:x + w] + mask * overlay_image
    # path = 'result'
    path = ''
    cv2.imwrite(os.path.join(path, f'1.png'), background)
    num_2 += 1
    print('插入成功。')
    init()

每次执行需要初始化x,y(图片匹配初始位置参数)，w,h（图片缩放初始宽高）

x = 0
y = 0
w = 0
h = 0
flag = True
threshold = 0
template = ''
num_1 = 0
num_2 = 0
photo3 = ''
"""参数初始化"""
def init():
    global x, y, w, h, threshold, template,flag
    x = 0
    y = 0
    w = 0
    h = 0
    threshold = 0
    template = ''

完整代码

import cv2
import datetime
import os
import numpy as np
x = 0
y = 0
w = 0
h = 0
flag = True
threshold = 0
template = ''
num_1 = 0
num_2 = 0
photo3 = ''
"""参数初始化"""
def init():
    global x, y, w, h, threshold, template,flag
    x = 0
    y = 0
    w = 0
    h = 0
    threshold = 0
    template = ''

"""检查模板图片中是否包含目标图片"""
def make_cv2(photo1, photo2):
    global x, y, w, h, num_1,flag
    starttime = datetime.datetime.now()
    img_rgb = cv2.imread(f'{photo1}')
    template = cv2.imread(f'{photo2}')
    h, w = template.shape[:-1]
    print('初始宽高', h, w)
    res = cv2.matchTemplate(img_rgb, template, cv2.TM_CCOEFF_NORMED)
    print('初始最大相似度', res.max())
    threshold = res.max()
    """，相似度小于0.2的，不予考虑；相似度在[0.2-0.75]之间的，逐渐缩小图片"""
    print(threshold)
    while threshold >= 0.1 and threshold <= 0.83:
        if w >= 20 and h >= 20:
            w = w - 1
            h = h - 1
            template = cv2.resize(
                template, (w, h), interpolation=cv2.INTER_CUBIC)
            res = cv2.matchTemplate(img_rgb, template, cv2.TM_CCOEFF_NORMED)
            threshold = res.max()
            print('宽度:', w, '高度:', h, '相似度:', threshold)
        else:
            break
    """达到0.75覆盖之前的图片"""
    if threshold > 0.8:
        loc = np.where(res >= threshold)
        x = int(loc[1])
        y = int(loc[0])
        print('覆盖图片左上角坐标:', x, y)
        for pt in zip(*loc[::-1]):
            cv2.rectangle(
                img_rgb, pt, (pt[0] + w, pt[1] + h), (255, 144, 51), 1)
        num_1 += 1
        endtime = datetime.datetime.now()
        print("耗时:", endtime - starttime)
        overlay_transparent(x, y, photo1, photo3)
    else:
        flag = False


"""将目标图片镶嵌到指定坐标位置"""
def overlay_transparent(x, y, photo1, photo3):
    y += 0
    global w, h, num_2
    background = cv2.imread(f'{photo1}')
    overlay = cv2.imread(f'{photo3}')
    """缩放图片大小"""
    overlay = cv2.resize(overlay, (w, h), interpolation=cv2.INTER_CUBIC)
    background_width = background.shape[1]
    background_height = background.shape[0]
    if x >= background_width or y >= background_height:
        return background
    h, w = overlay.shape[0], overlay.shape[1]
    if x + w > background_width:
        w = background_width - x
        overlay = overlay[:, :w]
    if y + h > background_height:
        h = background_height - y
        overlay = overlay[:h]
    if overlay.shape[2] < 4:
        overlay = np.concatenate([overlay, np.ones((overlay.shape[0], overlay.shape[1], 1), dtype=overlay.dtype) * 255],axis=2,)
    overlay_image = overlay[..., :3]
    mask = overlay[..., 3:] / 255.0
    background[y:y + h,x:x + w] = (1.0 - mask) * background[y:y + h,x:x + w] + mask * overlay_image
    # path = 'result'
    path = ''
    cv2.imwrite(os.path.join(path, f'1.png'), background)
    num_2 += 1
    print('插入成功。')
    init()


if __name__ == "__main__":
    photo1 = "1.png"
    photo2 = "3.png"
    photo3 = "white.png"

    while flag == True:
        make_cv2(photo1, photo2)
        overlay_transparent(x, y, photo1, photo3)

执行结果：