Upload CV1

_posts/2024-03-18-计算机视觉---1.数元素数目.md

@@ -0,0 +1,375 @@
+---
+title: 计算机视觉 - 1.数元素数目
+date: 2024-03-18 09:44:08 +0800
+categories: [Python, CV]
+tags: [Python, CV]      # TAG names should always be lowercase
+---
+
+# 计算机视觉1 - 数元素数目
+
+## 0. 任务目标
+
+1. 统计每幅图中各化学元素的数目；
+2. 统计AI和Fe元素、Fe和P元素、AI和P元素重叠的数目，并且以图像的形式展示出两两元素的重叠情况；
+3. 统计AI、Fe和P元素三者重叠的数目，并以图像的形式展示出三种元素重叠情况。
+4. 无论原始图像中的化学元素是何种颜色，均可实现本目标。
+
+## 1. 实现任务
+
+计算机视觉常用的方法是使用OpenCV的库函数对图像进行处理。因此我们先利用 `cv2.imread()` 函数读入图像数据。本例子中，共拥有3张图片，分别是Al，Fe和P（即铝、铁、磷）三种化学元素。下面是三张照片的样子。
+
+<img src="D:\Coder\Github Repos\StandardL.github.io\assets\img\posts\2024-03-18-计算机视觉1\Al.jpg" alt="Al" style="zoom:50%;" />
+
+<img src="D:\Coder\Github Repos\StandardL.github.io\assets\img\posts\2024-03-18-计算机视觉1\Fe.jpg" alt="Fe" style="zoom:50%;" />
+
+<img src="D:\Coder\Github Repos\StandardL.github.io\assets\img\posts\2024-03-18-计算机视觉1\P.jpg" alt="P" style="zoom:50%;" />
+
+读取的代码如下：
+
+```python
+# Set paths
+al_path = os.path.join('image', 'Al.jpg')
+fe_path = os.path.join('image', 'Fe.jpg')
+p_path = os.path.join('image', 'P.jpg')
+
+# Read images
+al = cv2.imread(al_path)
+fe = cv2.imread(fe_path)
+p = cv2.imread(p_path)
+```
+
+注意，这里的路径最好不要直接写死，而是用python os库中提供的 `os.path.join()` 函数进行拼接处理。该函数能够自行判断当前运行环境而选择路径的分隔符，无论是Linux还是Windows都不会出错。
+
+---
+
+读入图片后，我们发现图片中的左上角logo处和左下角的 50nm 标识都不是我们需要的东西，需要剔除。因此我们需要对输入图像进行预处理：
+
+- 为了去除左上角的logo，根据观察可以看到，logo读取进来的图像矩阵所在的行数均在74行及以内，因此我们处理的时候可以针对logo所在的行数进行删除。
+
+- 为了去除左下角的 50nm 标识，我们需要识别出白色区域并去除。
+
+为了实现上面的目标，我们先将图片从RGB色彩通道转换为HSV(HSB)色彩通道。OpenCV的库函数也提供了相应的转换。我们平常所见的图片绝大多数都是RGB通道，即利用Red（红）、Green（绿）、Blue（蓝）三种颜色及其浓度来表示一张彩色图片。在8Bit色深的RGB图像中，RGB三个通道的取值区间均为[0, 255]。而HSV色彩通道是利用Hue（色调）、Saturation（饱和度）、Value（明度）三值来表示颜色。不同于RGB图像，HSV图像是根据色调H来确定颜色的，而RGB图像一般需要通过三个值共同参与来表达一种颜色。
+
+处理代码如下：
+
+```python
+# RGB TO HSV
+def rgb2hsv(img):
+    hsv_img = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
+    # split H, S, V
+    h, s, v = cv2.split(hsv_img)
+    return [h, s, v]
+```
+
+显示HSV图像三通道的代码如下：
+
+```python
+def show_hsv_image(imgs: list):
+    cv2.imshow('H', imgs[0])
+    cv2.imshow('S', imgs[1])
+    cv2.imshow('V', imgs[2])
+    cv2.waitKey(0)
+    cv2.destroyAllWindows()
+```
+
+未经处理前，Al的分拆HSV通道后的图像如下：
+
+![Al的HSV图像](D:\Coder\Github Repos\StandardL.github.io\assets\img\posts\2024-03-18-计算机视觉1\Al的HSV图像.png)
+
+在HSV图像中，查询可知黑色点范围如下：
+
+- H: [0, 180]
+- S: [0, 255]
+- V: [0, 46]
+
+白色点范围如下：
+
+- H: [0, 180]
+- S: [0, 30]
+- V: [221, 255]
+
+logo区域的像素点较为明亮，因此可以直接利用位置和V的值进行判断去除。
+
+图像预处理代码如下：
+
+```python
+def remove_light(split_img: list):
+    h, s, v = split_img
+    for i in range(h.shape[0]):
+        for j in range(h.shape[1]):
+            # Remove black points
+            if 0 <= h[i, j] <= 180 and 0 <= v[i, j] <= 46:
+                h[i, j] = 0
+                s[i, j] = 0
+                v[i, j] = 0
+            # Remove white points
+            elif 0 <= h[i, j] <= 180 and 221 <= v[i, j] <= 255 and 0 <= s[i, j] <= 30:
+                h[i, j] = 0
+                s[i, j] = 0
+                v[i, j] = 0
+            # Remove logo area
+            if i <= 70 and 180 <= v[i, j] <= 255:
+                h[i, j] = 0
+                s[i, j] = 0
+                v[i, j] = 0
+
+    return [h, s, v]
+```
+
+---
+
+统计元素量时，为了方便统计，我们把有颜色的像素点都统计为1个元素。而我们已在上一步去除了杂色点，因此这一步可以直接数图像V通道的点数进行统计。统计函数如下：
+
+```python
+# Counting
+def count_points(img: list):
+    h, s, v = img
+    counter = 0
+    for i in range(v.shape[0]):
+        for j in range(v.shape[1]):
+            if v[i, j] != 0:
+                counter += 1
+
+    return counter
+```
+
+---
+
+统计重叠量时，依旧是利用上一步的思想进行统计，但是这时候我们需要两张图片的交集，即需要多加一个“与”的判断步骤。为了显示图片，我们还需要返回合并后的HSV矩阵。
+
+```python
+def count2overlap(img1: list, img2: list):
+    h1, s1, v1 = img1
+    h2, s2, v2 = img2
+    h0 = np.zeros(h1.shape, dtype=np.uint8)
+    s0 = np.zeros(s1.shape, dtype=np.uint8)
+    v0 = np.zeros(v1.shape, dtype=np.uint8)
+    counter = 0
+    for i in range(v1.shape[0]):
+        for j in range(v1.shape[1]):
+            if v1[i, j] != 0 and v2[i, j] != 0:
+                h0[i, j] = h1[i, j] + h2[i, j]
+                s0[i, j] = s1[i, j] + s2[i, j]
+                v0[i, j] = v1[i, j] + v2[i, j]
+                counter += 1
+
+    return counter, [h0, s0, v0]
+```
+
+三张图片的元素重叠量也是同理。
+
+```python
+def count3overlap(img1: list, img2: list, img3: list):
+    h1, s1, v1 = img1
+    h2, s2, v2 = img2
+    h3, s3, v3 = img3
+    h0 = np.zeros(h1.shape, dtype=np.uint8)
+    s0 = np.zeros(s1.shape, dtype=np.uint8)
+    v0 = np.zeros(v1.shape, dtype=np.uint8)
+    counter = 0
+    for i in range(v1.shape[0]):
+        for j in range(v1.shape[1]):
+            if v1[i, j] != 0 and v2[i, j] != 0 and v3[i, j] != 0:
+                h0[i, j] = h1[i, j] + h2[i, j] + h3[i, j]
+                s0[i, j] = s1[i, j] + s2[i, j] + s3[i, j]
+                v0[i, j] = v1[i, j] + v2[i, j] + v3[i, j]
+                counter += 1
+
+    return counter, [h0, s0, v0]
+```
+
+至此，我们本例子的目标基本完成。由于我们将重要的功能封装成函数且与具体颜色无关，因此无论原始图像中的化学元素是何种颜色，只需修改图像路径，均可实现本目标。
+
+---
+
+完整代码如下：
+
+```python
+import os.path
+import cv2
+import numpy as np
+import matplotlib.pyplot as plt
+
+# Set paths
+al_path = os.path.join('image', 'Al.jpg')
+fe_path = os.path.join('image', 'Fe.jpg')
+p_path = os.path.join('image', 'P.jpg')
+
+# Read images
+al = cv2.imread(al_path)
+fe = cv2.imread(fe_path)
+p = cv2.imread(p_path)
+
+
+# RGB TO HSV
+def rgb2hsv(img):
+    hsv_img = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
+    # split H, S, V
+    h, s, v = cv2.split(hsv_img)
+    return [h, s, v]
+
+
+# Plot image
+def show_image3(imgs: list):
+    plt.figure(figsize=(15, 15))
+    for i, img in enumerate(imgs):
+        plt.subplot(1, 3, i + 1)
+        plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
+        plt.axis('off')
+    plt.show()
+
+
+# Plot HSV image
+def show_hsv_image(imgs: list):
+    cv2.imshow('H', imgs[0])
+    cv2.imshow('S', imgs[1])
+    cv2.imshow('V', imgs[2])
+    cv2.waitKey(0)
+    cv2.destroyAllWindows()
+
+
+# Remove black and white points and logo
+def remove_light(split_img: list):
+    h, s, v = split_img
+    for i in range(h.shape[0]):
+        for j in range(h.shape[1]):
+            # Remove black points
+            if 0 <= h[i, j] <= 180 and 0 <= v[i, j] <= 46:
+                h[i, j] = 0
+                s[i, j] = 0
+                v[i, j] = 0
+            # Remove white points
+            elif 0 <= h[i, j] <= 180 and 221 <= v[i, j] <= 255 and 0 <= s[i, j] <= 30:
+                h[i, j] = 0
+                s[i, j] = 0
+                v[i, j] = 0
+            # Remove logo area
+            if i <= 70 and 180 <= v[i, j] <= 255:
+                h[i, j] = 0
+                s[i, j] = 0
+                v[i, j] = 0
+
+    return [h, s, v]
+
+
+# Counting
+def count_points(img: list):
+    h, s, v = img
+    counter = 0
+    for i in range(v.shape[0]):
+        for j in range(v.shape[1]):
+            if v[i, j] != 0:
+                counter += 1
+
+    return counter
+
+
+images = [al, fe, p]
+# show_image3(images)
+
+hsv_al = rgb2hsv(al)
+hsv_fe = rgb2hsv(fe)
+hsv_p = rgb2hsv(p)
+show_hsv_image(hsv_al)
+
+# pre-process
+processed_al = remove_light(hsv_al)
+processed_fe = remove_light(hsv_fe)
+processed_p = remove_light(hsv_p)
+
+# counting
+counter_al = count_points(processed_al)
+counter_fe = count_points(processed_fe)
+counter_p = count_points(processed_p)
+
+print("Al元素的点数：", counter_al)
+print("Fe元素的点数：", counter_fe)
+print("P元素的点数：", counter_p)
+
+
+# 求解重叠量，并将两者的重叠量画出来
+def count2overlap(img1: list, img2: list):
+    h1, s1, v1 = img1
+    h2, s2, v2 = img2
+    h0 = np.zeros(h1.shape, dtype=np.uint8)
+    s0 = np.zeros(s1.shape, dtype=np.uint8)
+    v0 = np.zeros(v1.shape, dtype=np.uint8)
+    counter = 0
+    for i in range(v1.shape[0]):
+        for j in range(v1.shape[1]):
+            if v1[i, j] != 0 and v2[i, j] != 0:
+                h0[i, j] = h1[i, j] + h2[i, j]
+                s0[i, j] = s1[i, j] + s2[i, j]
+                v0[i, j] = v1[i, j] + v2[i, j]
+                counter += 1
+
+    return counter, [h0, s0, v0]
+
+
+# 求解三者的重叠量
+def count3overlap(img1: list, img2: list, img3: list):
+    h1, s1, v1 = img1
+    h2, s2, v2 = img2
+    h3, s3, v3 = img3
+    h0 = np.zeros(h1.shape, dtype=np.uint8)
+    s0 = np.zeros(s1.shape, dtype=np.uint8)
+    v0 = np.zeros(v1.shape, dtype=np.uint8)
+    counter = 0
+    for i in range(v1.shape[0]):
+        for j in range(v1.shape[1]):
+            if v1[i, j] != 0 and v2[i, j] != 0 and v3[i, j] != 0:
+                h0[i, j] = h1[i, j] + h2[i, j] + h3[i, j]
+                s0[i, j] = s1[i, j] + s2[i, j] + s3[i, j]
+                v0[i, j] = v1[i, j] + v2[i, j] + v3[i, j]
+                counter += 1
+
+    return counter, [h0, s0, v0]
+
+
+# Get 2 overlap
+overlap_al_fe, overlap_img_af = count2overlap(processed_al, processed_fe)
+overlap_al_p, overlap_img_ap = count2overlap(processed_al, processed_p)
+overlap_fe_p, overlap_img_fp = count2overlap(processed_fe, processed_p)
+overlap_img_af = cv2.cvtColor(cv2.merge(overlap_img_af), cv2.COLOR_HSV2BGR)
+overlap_img_ap = cv2.cvtColor(cv2.merge(overlap_img_ap), cv2.COLOR_HSV2BGR)
+overlap_img_fp = cv2.cvtColor(cv2.merge(overlap_img_fp), cv2.COLOR_HSV2BGR)
+print("Al和Fe的重叠量：", overlap_al_fe)
+print("Al和P的重叠量：", overlap_al_p)
+print("Fe和P的重叠量：", overlap_fe_p)
+# show_image3([overlap_img_af, overlap_img_ap, overlap_img_fp])
+
+# Get 3 overlap
+overlap_all, overlap_img_all = count3overlap(processed_al, processed_fe, processed_p)
+overlap_img_all = cv2.cvtColor(cv2.merge(overlap_img_all), cv2.COLOR_HSV2BGR)
+print("Al, Fe和P的重叠量：", overlap_all)
+# cv2.imshow('Overlay with Al, Fe and P elements', overlap_img_all)
+# cv2.waitKey(0)
+# cv2.destroyAllWindows()
+
+# Save images
+if not os.path.exists('image_out'):
+    os.makedirs('image_out')
+cv2.imwrite(os.path.join('image_out', 'overlap_img_af.jpg'), overlap_img_af)
+cv2.imwrite(os.path.join('image_out', 'overlap_img_ap.jpg'), overlap_img_ap)
+cv2.imwrite(os.path.join('image_out', 'overlap_img_fp.jpg'), overlap_img_fp)
+cv2.imwrite(os.path.join('image_out', 'overlap_img_all.jpg'), overlap_img_all)
+
+```
+
+最终的得到的重叠图片如下：
+
+- Al和Fe：
+
+  <img src="D:\Coder\Github Repos\StandardL.github.io\assets\img\posts\2024-03-18-计算机视觉1\overlap_img_af.jpg" alt="overlap_img_af" style="zoom:50%;" />
+
+- Al和P：
+
+  <img src="D:\Coder\Github Repos\StandardL.github.io\assets\img\posts\2024-03-18-计算机视觉1\overlap_img_ap.jpg" alt="overlap_img_ap" style="zoom:50%;" />
+
+- Fe和P：
+
+  <img src="D:\Coder\Github Repos\StandardL.github.io\assets\img\posts\2024-03-18-计算机视觉1\overlap_img_fp.jpg" alt="overlap_img_fp" style="zoom:50%;" />
+
+- 三者重叠：
+
+  <img src="D:\Coder\Github Repos\StandardL.github.io\assets\img\posts\2024-03-18-计算机视觉1\overlap_img_all.jpg" alt="overlap_img_all" style="zoom:50%;" />