convolve and fft

最新推荐文章于 2024-09-05 09:01:35 发布

whuawell

最新推荐文章于 2024-09-05 09:01:35 发布

阅读量1k

点赞数

分类专栏： interest

本文链接：https://blog.csdn.net/whuawell/article/details/71188538

版权

interest 专栏收录该内容

6 篇文章 0 订阅

订阅专栏

mark

1. 时域卷积等同于频域乘积

2. gauss blur卷积等同于热传导方程偏微分方程演化过程

import numpy as np

def fft_convolve(a,b):
    n = len(a)+len(b)-1
    N = 2**(int(np.log2(n))+1)
    A = np.fft.fft(a, N)
    B = np.fft.fft(b, N)
    return np.fft.ifft(A*B)[:len(a)]

def fft_convolve2d(a,b,la, lb):

    n = la + lb - 1
    N = 2**(int(np.log2(n))+1)
    A = np.fft.fft2(a, [N,N])
    B = np.fft.fft2(b, [N,N])
    shift = (lb-1)/2
    return np.fft.ifft2(A*B)[shift:N-shift-1,shift:N-shift-1]

a = [x for x in range(1,6)]
b = [0.5 , 0.5]
c = np.convolve(a,b,mode='same')
print 'c=',c
print 'cf=', np.real(fft_convolve(a,b))

from scipy import signal as sg
from PIL import Image

def np_from_img(fname):
	img = Image.open(fname)
	img = img.convert('L')
	return np.asarray(img, dtype=np.float32)

def save_as_img(arr, fname):
	Image.fromarray(arr.round().astype(np.uint8)).save(fname)

def norm(arr):
	return 255.0 * np.absolute(arr)/np.max(arr)

gauss_temp = [[1,2,1],[2,4,2],[1,2,1]]
gauss_temp = np.array(gauss_temp)
A = np.random.rand(5,5)
#print 'A=',A
conv = sg.convolve2d(A, gauss_temp, 'same')
print conv
print '########################'
convfft = fft_convolve2d(A,gauss_temp,5,3)
print np.real(convfft)


# img = np_from_img('e:/lena.jpg')
# print img.shape
# save_as_img(norm(sg.convolve(img, [[1.],
#                                    [-1.]])),
#             'e:/lena-h.png')
# save_as_img(norm(sg.convolve(img, [[1., -1.]])),
#             'e:/lena-v.png')



# def common_size(a1, a2):
#     """Chop-off the first rows and cols from the two numpy arrays a1
#     and a2 so that they end up having the same size.
#     >>> print common_size(np.array([[0, 0],
#     ...                             [1, 2],
#     ...                             [3, 4]]),
#     ...                   np.array([[0, 5, 6],
#     ...                             [0, 7, 8]]))
#     (array([[1, 2],
#            [3, 4]]), array([[5, 6],
#            [7, 8]]))
#     """
#     (r1, c1) = a1.shape
#     (r2, c2) = a2.shape
#     return (a1[r1-r2 if r1>r2 else 0:,
#                c1-c2 if c1>c2 else 0:],
#             a2[r2-r1 if r2>r1 else 0::,
#                c2-c1 if c2>c1 else 0:])

# def gradient(im):
#     imv, imh = common_size(sg.convolve(im, [[1., -1.]]),
#                            sg.convolve(im, [[1.], [-1.]]))
#     return np.sqrt(np.power(imv, 2)+np.power(imh, 2))

# img = gradient(img)
# print img.shape
# save_as_img(img,'e:/gr.png')

from scipy import misc
import matplotlib.pyplot as plt

lena = misc.lena()
#save_as_img(lena, 'e:/lena-gray.png')

gauss_temp = [[1, 4, 7, 4, 1],
                          [4, 16, 26, 16, 4],   
                          [7, 26, 41, 26, 7],  
                          [4, 16, 26, 16, 4],   
                          [1, 4, 7, 4, 1] ]
lena_edge = sg.convolve2d(lena, gauss_temp, mode='same')
lena_edge = norm(lena_edge)
lena_edge = sg.convolve2d(lena_edge, gauss_temp, mode='same')
lena_edge = norm(lena_edge)
lena_edge = sg.convolve2d(lena_edge, gauss_temp, mode='same')
lena_edge = norm(lena_edge)
lena_edge = sg.convolve2d(lena_edge, gauss_temp, mode='same')
lena_edge = norm(lena_edge)
save_as_img(norm(lena_edge), 'e:/lena-edge.png')