# 失真






刚性变换： 只有物体的位置(平移变换)和朝向(旋转变换)发生改变，而形状不变，得到的变换称为刚性变换。刚性变换是最一般的变换。

使用透视变换，文字会扭曲失真。刚性变换就不会。

一些介绍： [https://blog.csdn.net/liuweiyuxiang/article/details/86510191](https://blog.csdn.net/liuweiyuxiang/article/details/86510191)

![在这里插入图片描述](/static/img/images/04473.png)

# Pillow的实现

创建一张空图写文字：

![在这里插入图片描述](/static/img/images/04474.png)

旋转了-30度：

![在这里插入图片描述](/static/img/images/04475.png)

将RGBA图像paste到大图中。

![在这里插入图片描述](/static/img/images/04476.png)

Python代码：

```python
from PIL import Image, ImageDraw, ImageFont
def draw_text_with_rotation(image, position, text, font_size, color, angle):
    draw = ImageDraw.Draw(image)
    font = ImageFont.truetype("STKAITI.TTF", font_size)
    width, height = draw.textsize(text, font=font)
    rotated_text = Image.new("RGBA", (width, height), (0, 0, 0, 0))
    text_draw = ImageDraw.Draw(rotated_text)
    text_draw.text((0, 0), text, font=font, fill=color)
    rotated_text.save("rotated_text1.png")
    rotated_text = rotated_text.rotate(-angle, resample=Image.BICUBIC, expand=True)
    rotated_text.save("rotated_text2.png")
    image.paste(rotated_text, position, rotated_text)
    rotated_text_width, rotated_text_height = rotated_text.size
    return rotated_text_width, rotated_text_height
# 创建一个空白图像
width, height = 800, 600
background_color = (255, 255, 255)  # 白色
image = Image.new("RGB", (width, height), background_color)
# 在大图中写入文本并旋转
text = "Hello, World!是吗可以的呀"
text_position = (200, 200)
text_font_size = 30
text_color = (0, 0, 0)  # 黑色
rotation_angle = -30
rotated_text_width, rotated_text_height = draw_text_with_rotation(image, text_position, text, text_font_size,
                                                                  text_color, rotation_angle)
# 画一个红色框框选上文字区域
draw = ImageDraw.Draw(image)
draw.rectangle((text_position[0], text_position[1], text_position[0] + rotated_text_width,
                text_position[1] + rotated_text_height), outline=(255, 0, 0))
# 保存图片
image.save("rotated_text3.png")
```

# Opencv的实现

## 不裁剪的旋转图像

参考imutils得到：

```python
import math
import cv2
import numpy as np
def rotate_bound2(image, angle):
    # grab the dimensions of the image and then determine the center
    # 抓取图像的尺寸，然后确定中心
    (h, w) = image.shape[:2]
    (cX, cY) = (w / 2, h / 2)
    # grab the rotation matrix (applying the negative of the angle to rotate clockwise), then grab the sine and cosine (i.e., the rotation components of the matrix)
    # 抓取旋转矩阵（应用角度的负数顺时针旋转），然后抓取正弦和余弦（即矩阵的旋转分量）
    M = cv2.getRotationMatrix2D((cX, cY), -angle, 1.0)
    cos = np.abs(M[0, 0])
    sin = np.abs(M[0, 1])
    # compute the new bounding dimensions of the image
    # 计算图像的新边界尺寸
    nW = int((h * sin) + (w * cos))
    nH = int((h * cos) + (w * sin))
    # adjust the rotation matrix to take into account translation
    # 调整旋转矩阵以考虑平移
    M[0, 2] += (nW / 2) - cX
    M[1, 2] += (nH / 2) - cY
    # perform the actual rotation and return the image
    # 执行实际旋转并返回图像
    return cv2.warpAffine(image, M, (nW, nH), flags=cv2.INTER_CUBIC)
def rotate_image(image, angle):
    angle = angle % 360.0
    if angle == 0:
        return image.copy()
    if angle == 180:
        return cv2.rotate(image, cv2.ROTATE_180)
    if angle == 90:
        return cv2.rotate(image, cv2.ROTATE_90_CLOCKWISE)
    if angle == 270:
        return cv2.rotate(image, cv2.ROTATE_90_COUNTERCLOCKWISE)
    return rotate_bound2(image, angle)
# 读取RGBA
image = cv2.imread('rotated_text1.png', cv2.IMREAD_UNCHANGED)
# 对图像进行旋转并扩展
rotated_image = rotate_image(image, 45)
# SAVE
cv2.imwrite("rotated_text1_rotated.png", rotated_image)
```

![在这里插入图片描述](/static/img/images/04477.png)

## 旋转文字并贴图

```python
import math
import cv2
import numpy as np
def rotate_bound2(image, angle):
    # grab the dimensions of the image and then determine the center
    # 抓取图像的尺寸，然后确定中心
    (h, w) = image.shape[:2]
    (cX, cY) = (w / 2, h / 2)
    # grab the rotation matrix (applying the negative of the angle to rotate clockwise), then grab the sine and cosine (i.e., the rotation components of the matrix)
    # 抓取旋转矩阵（应用角度的负数顺时针旋转），然后抓取正弦和余弦（即矩阵的旋转分量）
    M = cv2.getRotationMatrix2D((cX, cY), -angle, 1.0)
    cos = np.abs(M[0, 0])
    sin = np.abs(M[0, 1])
    # compute the new bounding dimensions of the image
    # 计算图像的新边界尺寸
    nW = int((h * sin) + (w * cos))
    nH = int((h * cos) + (w * sin))
    # adjust the rotation matrix to take into account translation
    # 调整旋转矩阵以考虑平移
    M[0, 2] += (nW / 2) - cX
    M[1, 2] += (nH / 2) - cY
    # perform the actual rotation and return the image
    # 执行实际旋转并返回图像
    return cv2.warpAffine(image, M, (nW, nH), flags=cv2.INTER_CUBIC)
def rotate_image(image, angle):
    angle = angle % 360.0
    if angle == 0:
        return image.copy()
    if angle == 180:
        return cv2.rotate(image, cv2.ROTATE_180)
    if angle == 90:
        return cv2.rotate(image, cv2.ROTATE_90_CLOCKWISE)
    if angle == 270:
        return cv2.rotate(image, cv2.ROTATE_90_COUNTERCLOCKWISE)
    return rotate_bound2(image, angle)
def draw_text_with_rotation(image, position, text, font_size, color, angle):
    # 定义字体样式
    font = cv2.FONT_HERSHEY_SIMPLEX
    thickness = 2
    # 获取文字的宽度和高度
    text_size, baseline = cv2.getTextSize(text, font, font_size, thickness)
    print(text_size, baseline)
    text_width, text_height = text_size
    # 灰度图上白纸黑字
    text_image = np.zeros((text_height + baseline, text_width, 1), np.uint8)
    cv2.putText(text_image, text, (0, text_height), font, font_size, (255,), thickness)  # top-left corner.
    # 灰度图给到透明通道上
    text_image2 = np.zeros((text_height + baseline, text_width, 4), np.uint8)
    text_image2[:, :, 3] = text_image[:, :, 0]
    text_image = text_image2
    cv2.imwrite("self1.png", text_image)
    rotated_text = rotate_image(text_image, angle)
    cv2.imwrite("self2.png", rotated_text)
    rotated_text_height, rotated_text_width = rotated_text.shape[:2]
    image[position[0]:position[0] + rotated_text_height, position[1]:position[1] + rotated_text_width] = rotated_text
    return rotated_text_width, rotated_text_height
# 创建一个空白图像
width, height = 800, 600
image = np.zeros((height, width, 4), np.uint8)
# 在大图中写入文本并旋转
text = "Hello, World"
text_position = (200, 200)
text_font_size = 1
text_color = (255, 255, 255)  # 黑色
rotation_angle = -30
rotated_text_width, rotated_text_height = draw_text_with_rotation(image, text_position, text, text_font_size,
                                                                  text_color, rotation_angle)
# 画一个红色框框选上文字区域
cv2.rectangle(image, (text_position[0], text_position[1]),
              (text_position[0] + rotated_text_width, text_position[1] + rotated_text_height), (0, 0, 255, 255), 2)
# 保存图片
cv2.imwrite("self3.png", image)
```

文字图：

![在这里插入图片描述](/static/img/images/04478.png)

旋转后：

![在这里插入图片描述](/static/img/images/04479.png)

贴图：

![在这里插入图片描述](/static/img/images/04480.png)




## C++的图片透视变换

```cpp
        // textImgMask的四个顶点
        std::vector<cv::Point> ptsSrc;
        ptsSrc.emplace_back(0, 0);
        ptsSrc.emplace_back(textImgMask.cols, 0);
        ptsSrc.emplace_back(textImgMask.cols, textImgMask.rows);
        ptsSrc.emplace_back(0, textImgMask.rows);
        // ptsImgText的四个顶点
        std::vector<cv::Point> ptsDst;
        ptsDst.emplace_back(ptsImgText[0]);
        ptsDst.emplace_back(ptsImgText[1]);
        ptsDst.emplace_back(ptsImgText[2]);
        ptsDst.emplace_back(ptsImgText[3]);
        // 计算透视变换矩阵
        cv::Mat homography = cv::findHomography(ptsSrc, ptsDst);
        // 透视变换到imgOutTmp
        cv::Mat imgOutTmp = cv::Mat::zeros(img.size(), CV_8UC3);
        cv::warpPerspective(textImgMask, imgOutTmp, homography, img.size());
        // 转到原图
        cv::Mat gray;
        cv::cvtColor(imgOutTmp, gray, cv::COLOR_BGR2GRAY);
        imgOut.setTo(cv::Scalar(textColors[i][2], textColors[i][1], textColors[i][0]), gray > 125);  // mask部分改色
```

【Opencv】PIL Opencv 向图片写入文字并旋转文字，Opencv图片旋转不截断，Opencv图片旋转不裁剪

版本不一样的时候，返回也不一样。





我使用opencv/4.5.5。

下图是使用minAreaRect判定的角度，可以看到，数值范围是[0,90]，看起来很离谱。

![在这里插入图片描述](/static/img/images/04636.jpeg)

画出这张图使用的程序如下：

C++
```cpp
int main() {
    std::string prefix1 =
        "/mnt/c/Users/dong.xie/Desktop/workcode/cpp_project/photo_translate/03implement/photo_translate/"
        "test_photo_translate/";
    // 写一段程序，显示opencv中的角度是怎么计算的
    // 构建一个全白色的4000*4000的图片
    cv::Mat img = cv::Mat::zeros(4000, 4000, CV_8UC3);
    img.setTo(cv::Scalar(114, 114, 114));
    // 遍历0度到360度
    for (int i = 0; i < 360; i = i + 10) {
        //(2000,2000)与(3200,2000)是一条直线，(2000,2000)是原点，不断旋转这条直线，看看旋转后的直线的角度是多少
        int p0x = 2000;
        int p0y = 2000;
        int p1x = 3200;
        int p1y = 2000;
        p1x = 1200 * std::cos(i * 3.1415926 / 180) + p0x;
        p1y = -1200 * std::sin(i * 3.1415926 / 180) + p0y;
        cv::Mat pts = (cv::Mat_<float>(4, 2) << p0x, p0y, p1x, p1y, p0x + 2, p0y + 2, p1x + 2, p1y + 2);
        // 绘制(p0x,p0y)与（p1x，p1y)这条直线
        cv::line(img, cv::Point2f(p0x, p0y), cv::Point2f(p1x, p1y), cv::Scalar(0, 0, 255), 2);
        cv::RotatedRect rect = cv::minAreaRect(pts);
        // 画出旋转后的直线
        cv::Point2f vertices[4];
        rect.points(vertices);
        for (int i = 0; i < 4; i++)
            cv::line(img, vertices[i], vertices[(i + 1) % 4], cv::Scalar(0, 0, 255), 2);
        // 显示旋转后的直线的角度
        cv::putText(img,
                    std::to_string(rect.angle).substr(0, 5),
                    cv::Point(p1x, p1y),
                    cv::FONT_HERSHEY_SIMPLEX,
                    2,
                    cv::Scalar(0, 255, 0),
                    2);
    }
    cv::imwrite(prefix1 + "test.jpg", img);
    return 0;
}
```

那么python中呢，如果版本是一样的，那么结果是一样的，如果版本不同，比如4.4，那么就不会是一样的结果。

```python
import cv2
import numpy as np
print(cv2.__version__)
img = np.zeros((4000, 4000, 3), np.uint8)
img[:] = 114
for i in range(0, 360, 10):
    p0x = 2000
    p0y = 2000
    p1x = 3200
    p1y = 2000
    p1x = 1200 * np.cos(i * np.pi / 180) + p0x
    p1y = -1200 * np.sin(i * np.pi / 180) + p0y
    pts = np.array([[p0x, p0y], [p1x, p1y], [p0x + 2, p0y + 2], [p1x + 2, p1y + 2]], np.float32)
    cv2.line(img, (int(p0x), int(p0y)), (int(p1x), int(p1y)), (0, 0, 255), 2)
    rect = cv2.minAreaRect(pts)
    box = cv2.boxPoints(rect)
    box = np.int0(box)
    cv2.drawContours(img, [box], 0, (0, 0, 255), 2)
    cv2.putText(img, str(rect[2])[0:5], (int(p1x), int(p1y)), cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 255, 0), 2)
cv2.imwrite('test.jpg', img)
```

![在这里插入图片描述](/static/img/images/04637.jpeg)

C++的OpenCV中cv::minAreaRect的返回角度的数值范围是多少？

# 传统方法





程序
```python
import numpy as np
import cv2
def replaceZeroes(data):
    min_nonzero = min(data[np.nonzero(data)])
    data[data == 0] = min_nonzero
    return data
def SSR(src_img, size):
    L_blur = cv2.GaussianBlur(src_img, (size, size), 0)
    img = replaceZeroes(src_img)
    L_blur = replaceZeroes(L_blur)
    dst_Img = cv2.log(img / 255.0)
    dst_Lblur = cv2.log(L_blur / 255.0)
    dst_IxL = cv2.multiply(dst_Img, dst_Lblur)
    log_R = cv2.subtract(dst_Img, dst_IxL)
    dst_R = cv2.normalize(log_R, None, 0, 255, cv2.NORM_MINMAX)
    log_uint8 = cv2.convertScaleAbs(dst_R)
    return log_uint8
# 使用照片修复的办法来去掉图片中的文字
def remove_text1(path):
    size = 3
    src_img = cv2.imread(path)
    # # 膨胀图片
    # src_img = cv2.dilate(src_img, kernel=np.ones((3, 3), dtype=np.uint8), iterations=2)
    # 腐蚀图片
    # src_img = cv2.erode(src_img, kernel=np.ones((3, 3), dtype=np.uint8), iterations=2)
    b_gray, g_gray, r_gray = cv2.split(src_img)
    b_gray = SSR(b_gray, size)
    g_gray = SSR(g_gray, size)
    r_gray = SSR(r_gray, size)
    ssrimg = cv2.merge([b_gray, g_gray, r_gray])
    canny_img = cv2.Canny(ssrimg, 200, 150)
    k = np.ones((3, 3), np.uint8)
    mask = cv2.morphologyEx(canny_img, cv2.MORPH_CLOSE, k)  # 闭运算
    # 对mask进行膨胀
    # mask = cv2.dilate(mask, kernel=np.ones((3, 3), dtype=np.uint8), iterations=6)
    dst = cv2.inpaint(src_img, mask, 3, cv2.INPAINT_TELEA)
    # dst = cv2.inpaint(dst, mask, 3, cv2.INPAINT_TELEA)
    # 膨胀
    dst = cv2.dilate(dst, kernel=np.ones((3, 3), dtype=np.uint8), iterations=5)
    return dst
# 使用mask直接填充颜色的办法来去掉图片中的文字
def remove_text2(path):
    size = 3
    src_img = cv2.imread(path)
    # 腐蚀图片
    # src_img = cv2.erode(src_img, kernel=np.ones((3, 3), dtype=np.uint8), iterations=2)
    b_gray, g_gray, r_gray = cv2.split(src_img)
    b_gray = SSR(b_gray, size)
    g_gray = SSR(g_gray, size)
    r_gray = SSR(r_gray, size)
    ssrimg = cv2.merge([b_gray, g_gray, r_gray])
    canny_img = cv2.Canny(ssrimg, 200, 150)
    k = np.ones((3, 3), np.uint8)
    mask = cv2.morphologyEx(canny_img, cv2.MORPH_CLOSE, k)  # 闭运算
    # 保存mask为图片
    cv2.imwrite("mask.png", mask)
    # mask等于1的区域，用白色填充
    # dst = cv2.bitwise_and(src_img, src_img, mask=mask)
    src_img[mask == 255] = (255, 255, 255)
    # 膨胀
    src_img = cv2.dilate(src_img, kernel=np.ones((3, 3), dtype=np.uint8), iterations=5)
    return src_img
# 移除背景色，增强文字
def remove_ground(path):
    size = 3
    src_img = cv2.imread(path)
    b_gray, g_gray, r_gray = cv2.split(src_img)
    b_gray = SSR(b_gray, size)
    g_gray = SSR(g_gray, size)
    r_gray = SSR(r_gray, size)
    ssrimg = cv2.merge([b_gray, g_gray, r_gray])
    canny_img = cv2.Canny(ssrimg, 200, 150)
    k = np.ones((3, 3), np.uint8)
    mask = cv2.morphologyEx(canny_img, cv2.MORPH_CLOSE, k)  # 闭运算
    # mask等于1的区域，用白色填充
    dst = cv2.bitwise_and(src_img, src_img, mask=mask)
    dst[mask == 0] = (255, 255, 255)
    return dst
if __name__ == '__main__':
    imgPath = "papertext.jpg"
    dst = remove_text2(imgPath)
    cv2.imwrite("result.jpg", dst)
```

原图：

![在这里插入图片描述](/static/img/images/04728.jpeg)

找到文字部分：

![在这里插入图片描述](/static/img/images/04729.png)

去除文字：

![在这里插入图片描述](/static/img/images/04730.jpeg)

保留清晰文字：

![在这里插入图片描述](/static/img/images/04731.jpeg)

# 深度学习

1、

Self-Supervised Text Erasing with Controllable Image Synthesis 即STE ： 论文[https://arxiv.org/abs/2204.12743](https://arxiv.org/abs/2204.12743)。

2、

EnsNet ： [https://blog.csdn.net/u013554402/article/details/104494261](https://blog.csdn.net/u013554402/article/details/104494261)。

3、

[https://paperswithcode.com/paper/mtrnet-one-stage-mask-based-scene-text-eraser](https://paperswithcode.com/paper/mtrnet-one-stage-mask-based-scene-text-eraser)。

4、

[http://www.cjig.cn/html/jig/2022/4/20220420.htm](http://www.cjig.cn/html/jig/2022/4/20220420.htm)





# 译文回填

SickZil-Machine用于去除文本： [https://github.com/KUR-creative/SickZil-Machine](https://github.com/KUR-creative/SickZil-Machine)

lama-cleaner技术：  [https://github.com/Sanster/lama-cleaner](https://github.com/Sanster/lama-cleaner)

【OpenCV】【深度学习】去除图片文字；只保留图片文字；译文回填

原因是png图片里面的profile ，好像又听说高版本的png在做什么。






解决方式很简单，重写一下这个图。


下面程序会读取 r"/ssd/xd/datasets/multilabelsTask/"下面的所有jpg和png图，多进程方式地重写一下图片，注意，如果opencv打不开这个图，这个图会被删除：
```python
import multiprocessing
import os
import cv2
import numpy as np
def listPathAllfiles(dirname):
    result = []
    for maindir, subdir, file_name_list in os.walk(dirname):
        for filename in file_name_list:
            apath = os.path.join(maindir, filename)
            result.append(apath)
    return result
def overwrite(files1):
    for fname in files1:
        try:
            if fname.endswith(".jpg"):
                img = cv2.imdecode(np.fromfile(fname, dtype=np.uint8), 1)  # img是矩阵
                cv2.imencode('.jpg', img)[1].tofile(fname)
            elif fname.endswith(".png"):
                img = cv2.imdecode(np.fromfile(fname, dtype=np.uint8), 1)  # img是矩阵
                cv2.imencode('.png', img)[1].tofile(fname)
        except:
            os.remove(fname)
            print(fname)
if __name__ == '__main__':
    root = r"/ssd/xd/datasets/multilabelsTask/"
    files = listPathAllfiles(root)
    p = multiprocessing.Pool()  # 创建一个包含2个进程的进程池
    # split files to several parts
    for i in range(0, len(files), 10000):
        p.apply_async(func=overwrite, args=(files[i:i + 1000],))  # 往池子里加一个异步执行的子进城
    p.close()  # 等子进程执行完毕后关闭进程池
    p.join()  # 主进程等待
```

【opencv】【python】libpng warning: iCCP: known incorrect sRGB profile 解决

截图目标，分析目标的颜色直方图：





```python
import cv2
from matplotlib import pyplot as plt
def show_img(path):
    img = cv2.imread(path)
    b, g, r = img[:, :, 0], img[:, :, 1], img[:, :, 2]
    hist_b = cv2.calcHist([b], [0], None, [256], [0, 256])
    hist_g = cv2.calcHist([g], [0], None, [256], [0, 256])
    hist_r = cv2.calcHist([r], [0], None, [256], [0, 256])
    plt.figure(figsize=(10, 5))
    plt.subplot(121)
    plt.plot(hist_r, color='r', label="r")
    plt.plot(hist_g, color='g', label="g")
    plt.plot(hist_b, color='b', label="b")
    plt.legend()
    img2 = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
    h, s, v = img2[:, :, 0], img2[:, :, 1], img2[:, :, 2]
    hist_h = cv2.calcHist([h], [0], None, [256], [0, 256])
    hist_s = cv2.calcHist([s], [0], None, [256], [0, 256])
    hist_v = cv2.calcHist([v], [0], None, [256], [0, 256])
    plt.subplot(122)
    plt.plot(hist_h, color='r', label="h")
    plt.plot(hist_s, color='g', label="s")
    plt.plot(hist_v, color='b', label="v")
    plt.legend()
    plt.show()
    return hist_r, hist_g, hist_b, hist_h, hist_s, hist_v
r, g, b, h, s, v = show_img("lanse.png")
```

![在这里插入图片描述](/static/img/images/00262.png)


```
import cv2
import numpy as np
from matplotlib import pyplot as plt
img = cv2.imread("1111.jpg")
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
lower_red = np.array([115, 0, 160])
upper_red = np.array([140, 30, 190])
mask = cv2.inRange(hsv, lower_red, upper_red)
res = cv2.bitwise_and(img, img, mask=mask)
# cv2.imshow('img', img)
cv2.imshow('mask', mask)
cv2.imshow('res', res)
cv2.waitKey(0)
cv2.destroyAllWindows()
```

Opencv 颜色过滤 Python mask 选取颜色 颜色直方图

我是一个专注技术分享的博主,尤其爱搞深度学习和单片机电子。你现在看到的是我的个人博客网站。