Python FFT实现

import numpy as np
import cv2
import matplotlib.pyplot as plt

# # file_path = 'D:\\code_python\\KAIR\\visualization\\x0.png'
# file_path = 'D:\\dataset\\test\\classic5\\lena.bmp'
# img = cv2.imread(file_path)[:,:,0] #cv2默认是BGR通道顺序，这里调整到RGB
# # img = cv2.resize(img,(500,500))

# fre = np.fft.fft2(img,axes=(0,1)) #变换得到的频域图数据是复数组成的
# fre_m = np.abs(fre)   #幅度谱，求模得到
# fre_p = np.angle(fre) #相位谱，求相角得到
# fre_m_real_log = np.log(fre_m)
# plt.figure()
# plt.imshow(fre_m_real_log.astype('uint8'))




# fre_shift = np.fft.fftshift(fre)
# fre_shift_real = np.abs(fre_shift)
# fre_shift_real_log = np.log(fre_shift_real)


# img2 = cv2.imread(file_path)[:,:,1] #cv2默认是BGR通道顺序，这里调整到RGB
# # img = cv2.resize(img,(500,500))
# fre2 = np.fft.fft2(img2,axes=(0,1)) #变换得到的频域图数据是复数组成的
# fre_shift2 = np.fft.fftshift(fre2)
# fre_shift_real2 = np.abs(fre_shift2)
# fre_shift_real_log2 = np.log(fre_shift_real2)



# #把振幅设为常数
# constant = fre_m.mean()
# fre_ = constant * np.e**(1j*fre_p)   #把幅度谱和相位谱再合并为复数形式的频域图数据
# img_onlyphase = np.abs(np.fft.ifft2(fre_,axes=(0,1))) #还原为空间域图像
# #把相位设为常数
# constant = fre_p.mean()
# fre_ = fre_m * np.e**(1j*constant) 
# img_onlymagnitude = np.abs(np.fft.ifft2(fre_,axes=(0,1)))

# plt.figure()
# plt.imshow(fre.astype('uint8'))
# plt.figure()
# plt.imshow(fre_shift_real_log.astype('uint8'))
# plt.figure()
# plt.imshow(fre_shift_real_log2.astype('uint8'))
# # plt.figure()
# # img_onlyphase_hist = plt.hist(img_onlyphase.astype('uint8').ravel())
# # plt.figure()
# # img_onlymagnitude_hist = plt.hist(img_onlymagnitude.astype('uint8').ravel())
# # plt.figure()
# # img_hist = plt.hist(img.astype('uint8').ravel())


# plt.figure()
# plt.imshow(img.astype('uint8'))
# # plt.figure()
# # plt.imshow(img_onlyphase.astype('uint8'))
# # plt.figure()
# # plt.imshow(img_onlymagnitude.astype('uint8'))
# # plt.figure()
# # plt.imshow(fre.astype('uint8'))
# # plt.figure()
# # plt.imshow(fre_m.astype('uint8'))
# # plt.figure()
# # plt.imshow(fre_p.astype('uint8'))
# # plt.figure()
# # plt.imshow(fre_.astype('uint8'))
# plt.show()



######### plot frequency curve
#### one image
import os
def plot_fre(path, legend_label):
    img = cv2.imread(path)
    # img = cv2.resize(img, (0,0), fx=0.5, fy=0.5)
    img_YCrCb = cv2.cvtColor(img, cv2.COLOR_BGR2YCrCb)
    img_Y = img_YCrCb[:,:,0] #cv2默认是BGR通道顺序，这里调整到RGB
    # img = cv2.resize(img,(500,500))

    fre = np.fft.fft2(img_Y,axes=(0,1)) #变换得到的频域图数据是复数组成的
    # fre = np.fft.fftshift(fre)
    fre_m = np.abs(fre)   #幅度谱，求模得到
    fre_p = np.angle(fre) #相位谱，求相角得到
    fre_m_real_log = np.log(fre_m)

    # x = np.linspace(0,128,128)
    # y = np.linspace(0,1,128)
    # xx, yy = np.meshgrid(x, y)
    # # plt.contourf(xx,yy,fre_m_real_log.astype('uint8'))
    # # plt.show()

    # plt.figure()
    # plt.imshow(fre_m_real_log.astype('uint8'))
    # plt.xticks([])
    # plt.yticks([])

    # plt.show()
    # ## only x
    # x = [_ for _ in range(fre_m_real_log.shape[-1]//2)]
    # y = [fre_m_real_log[0,i] for i in range(fre_m_real_log.shape[-1]//2)]
    # plt.figure()
    # plt.plot(x,y,label='only x')


    # ## x and y
    length = min(fre_m_real_log.shape[0]//2, fre_m_real_log.shape[1]//2)
    x = [_ for _ in range(length)]
    y = [(sum(fre_m_real_log[i,j] for j in range(i))+sum(fre_m_real_log[j,i] for j in range(i))+fre_m_real_log[i,i])/(2*i+1) for i in range(length)]
    # plt.figure()
    plt.plot(x,y,label=legend_label)
path1 = 'D:\dataset\SIDD_Valid\img\gt'
path2 = 'D:\dataset\SIDD_Valid\img\\noisy'
img_name = 'test_SIDD_702.png'
plot_fre(os.path.join(path1, img_name), img_name[:-4]+'_gt')
plot_fre(os.path.join(path2, img_name), img_name[:-4]+'_noisy')
plt.legend()
plt.show()

#### dataset
# import os
# dataset_path = 'D:\dataset\SIDD_Valid\img\\noisy'
# dataset_path_gt = 'D:\dataset\SIDD_Valid\img\gt'
# img_names = os.listdir(dataset_path)


# for ii, img_name in enumerate(img_names):
#     img_rgb = cv2.imread(os.path.join(dataset_path, img_name))
#     img_YCrCb = cv2.cvtColor(img_rgb, cv2.COLOR_BGR2YCrCb)
#     img_Y = img_YCrCb[:,:,0]

#     fre = np.fft.fft2(img_Y,axes=(0,1))
#     fre_m = np.abs(fre)   #幅度谱，求模得到
#     fre_p = np.angle(fre) #相位谱，求相角得到
#     fre_m_real_log = np.log(fre_m)

#     length = min(fre_m_real_log.shape[0]//2, fre_m_real_log.shape[1]//2)
#     x = [_ for _ in range(length)]
#     y = [(sum(fre_m_real_log[i,j] for j in range(i))+sum(fre_m_real_log[j,i] for j in range(i))+fre_m_real_log[i,i])/(2*i+1) for i in range(length)]
#     if ii==0:
#         y_sum=y
#     else:
#         y_sum = [y[i]+y_sum[i] for i in range(length)]
    

#     img_rgb_gt = cv2.imread(os.path.join(dataset_path_gt, img_name))
#     img_YCrCb_gt = cv2.cvtColor(img_rgb_gt, cv2.COLOR_BGR2YCrCb)
#     img_Y_gt = img_YCrCb_gt[:,:,0] 
    
#     fre_gt = np.fft.fft2(img_Y_gt,axes=(0,1))
#     fre_m_gt = np.abs(fre_gt)   #幅度谱，求模得到
#     fre_p_gt = np.angle(fre_gt) #相位谱，求相角得到
#     fre_m_real_log_gt = np.log(fre_m_gt)

#     length = min(fre_m_real_log_gt.shape[0]//2, fre_m_real_log_gt.shape[1]//2)
#     x_gt = [_ for _ in range(length)]
#     y_gt = [(sum(fre_m_real_log_gt[i,j] for j in range(i))+sum(fre_m_real_log_gt[j,i] for j in range(i))+fre_m_real_log_gt[i,i])/(2*i+1) for i in range(length)]
#     if ii==0:
#         y_sum_gt=y_gt
#     else:
#         y_sum_gt = [y_gt[i]+y_sum_gt[i] for i in range(length)]   


# y_ave = [y_sum[i]/(ii+1) for i in range(length)]
# plt.plot(x,y_ave,label='SIDD_noisy')

# y_ave_gt = [y_sum_gt[i]/(ii+1) for i in range(length)]
# plt.plot(x,y_ave_gt,label='SIDD_GT')
# plt.legend()
# plt.show()