【计算机视觉中的傅里叶变换】2. 实践丨1. 离散余弦变换-图像频率成分分析丨_python

待处理图像：Lena.jpg
开始分析
import numpy as np
import matplotlib.pyplot as plt
import cv2
import copy

#######################
###   1. 读取图片   ###
#######################
# image = cv2.imread('Lena.jpg', 0) 
image = cv2.imread('Lena.jpg')
(img_B, img_G, img_R) = cv2.split(image)
print(img_R.shape)

###########################################
###    2. 离散余弦变换，并获取其幅频谱   ###
###########################################
# 1. DCT
img_B_dct = cv2.dct(np.float32(img_B))
img_G_dct = cv2.dct(np.float32(img_G))
img_R_dct = cv2.dct(np.float32(img_R))

# 2. Frequency Component
img_B_dct_log = 20 * np.log(abs(img_B_dct))
img_G_dct_log = 20 * np.log(abs(img_G_dct))
img_R_dct_log = 20 * np.log(abs(img_R_dct))

# 3. iDCT
img_B_back = cv2.idct(img_B_dct)
img_G_back = cv2.idct(img_G_dct)
img_R_back = cv2.idct(img_R_dct)

######################
###    3. 可视化   ###
######################
plt.subplot(3, 3, 1)
plt.imshow(img_B, cmap='gray')
plt.title('B channel of Origin Img', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(3, 3, 2)
plt.imshow(img_B_dct_log, cmap='gray')
plt.title('DCT of B channel', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(3, 3, 3)
plt.imshow(img_B_back, cmap='gray')
plt.title('iDCT of B channel', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(3, 3, 4)
plt.imshow(img_G, cmap='gray')
plt.title('G channel of Origin Img', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(3, 3, 5)
plt.imshow(img_G_dct_log, cmap='gray')
plt.title('DCT of G channel', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(3, 3, 6)
plt.imshow(img_G_back, cmap='gray')
plt.title('iDCT of G channel', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(3, 3, 7)
plt.imshow(img_R, cmap='gray')
plt.title('R channel of Origin Img', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(3, 3, 8)
plt.imshow(img_R_dct_log, cmap='gray')
plt.title('DCT of R channel', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(3, 3, 9)
plt.imshow(img_R_back, cmap='gray')
plt.title('iDCT of R channel', fontproperties='Times New Roman')
plt.axis('off')

plt.savefig('1.png', dpi=300)
# plt.show()
plt.close()

###############################
###   4. 分析不同频率成分   ###
###############################

### 1. 准备
image_B_dct = copy.deepcopy(img_B_dct)
image_G_dct = copy.deepcopy(img_G_dct)
image_R_dct = copy.deepcopy(img_R_dct)

image_B_dct_log = 20*np.log(abs(image_B_dct))
image_G_dct_log = 20*np.log(abs(image_G_dct))
image_R_dct_log = 20*np.log(abs(image_R_dct))

### 2. 选择频率
LOW  = 25
HIGH = 100
# (1) B
for i in range(image_B_dct.shape[0]):
    for j in range(image_B_dct.shape[1]):
        if i < LOW and j < LOW:
            image_B_dct[i, j] = 0
            image_B_dct_log[i, j] = 0
        if i >= HIGH or j >= HIGH:
            image_B_dct[i, j] = 0
            image_B_dct_log[i, j] = 0
# (2) G
for i in range(image_G_dct.shape[0]):
    for j in range(image_G_dct.shape[1]):
        if i < LOW and j < LOW:
            image_G_dct[i, j] = 0
            image_G_dct_log[i, j] = 0
        if i >= HIGH or j >= HIGH:
            image_G_dct[i, j] = 0
            image_G_dct_log[i, j] = 0
# (3) R
for i in range(image_R_dct.shape[0]):
    for j in range(image_R_dct.shape[1]):
        if i < LOW and j < LOW:
            image_R_dct[i, j] = 0
            image_R_dct_log[i, j] = 0
        if i >= HIGH or j >= HIGH:
            image_R_dct[i, j] = 0
            image_R_dct_log[i, j] = 0

### 3. 逆DCT
img_B_back = cv2.idct(image_B_dct)
img_G_back = cv2.idct(image_G_dct)
img_R_back = cv2.idct(image_R_dct)

plt.subplot(2, 3, 1)
plt.imshow(image_B_dct_log, cmap='gray')
plt.title('compression B', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(2, 3, 4)
plt.imshow(img_B_back, cmap='gray')
plt.title('B back', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(2, 3, 2)
plt.imshow(image_G_dct_log, cmap='gray')
plt.title('compression G', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(2, 3, 5)
plt.imshow(img_G_back, cmap='gray')
plt.title('G Back', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(2, 3, 3)
plt.imshow(image_R_dct_log, cmap='gray')
plt.title('compression R', fontproperties='Times New Roman')
plt.axis('off')

plt.subplot(2, 3, 6)
plt.imshow(img_R_back, cmap='gray')
plt.title('R back', fontproperties='Times New Roman')
plt.axis('off')

plt.savefig('2.png', dpi=300)
# plt.show()
plt.close()

### 4. 合并通道
img_back = cv2.merge([img_B_back, img_G_back, img_R_back])
img_back = cv2.cvtColor(img_back, cv2.COLOR_BGR2RGB)

img_back = (img_back - img_back.min()) / (img_back.max()-img_back.min()) *255
img_back = img_back.astype('uint8')

plt.subplot(1,1,1)
plt.imshow(img_back)
plt.title('Combine back', fontproperties='Times New Roman')
plt.axis('off')

plt.savefig('3.png', dpi=300)
# plt.show()
plt.close()