小波变换使用matlab和python

注意1d和2d的函数区别。注意默认参数问题。最终三个版本结果能够对齐。

matlab

load('wave_in.mat')

% res: image of 1536 x 1536
th=1;
dlevel=7;
wavename='db6';

[m,n] = wavedec2(res, dlevel, wavename);

vec = zeros(size(m));
vec(1:n(1)*n(1)*1) = m(1:n(1)*n(1)*1);

background =  waverec2(vec, n, wavename);
background(background<0.001)=0;

python version1

python">import time
import torch
from mat_utils import load_mat
import numpy as np
import pywt
import copy

data_dict = load_mat("test_data/wave_in.mat")
res = data_dict["res"]
# res: image of 1536 x 1536

data_torch = torch.from_numpy(res)

wavelet = pywt.Wavelet('db6')

wavename = 'db6'
dlevel = 7

coeffs_py = pywt.wavedec2(res, wavename, level=dlevel)

# keep fist one and zero all others
for i in range(1, len(coeffs_py)):
    sub_coefs = list(coeffs_py[i])
    sub_coefs = [np.zeros_like(tensor) for tensor in sub_coefs]
    coeffs_py[i] = sub_coefs

background_py = pywt.waverec2(coeffs_py, wavename)

python version2 with pytorch

使用库

https://github.com/v0lta/PyTorch-Wavelet-Toolbox

python">import time
import torch
from mat_utils import load_mat
import numpy as np
import pywt
import ptwt
# pip install ptwt

data_dict = load_mat("test_data/wave_in.mat")
res = data_dict["res"]
# res: image of 1536 x 1536

data_torch = torch.from_numpy(res)

wavename = 'db6'
dlevel = 7

wavelet = pywt.Wavelet(wavename)

coeffs_pt = ptwt.wavedec2(data_torch, wavelet, level=dlevel, mode='symmetric')
coeffs_pt = list(coeffs_pt)

# keep fist one and zero all others
for i in range(1, len(coeffs_pt)):
    sub_coefs = list(coeffs_pt[i])
    sub_coefs = [torch.zeros_like(tensor) for tensor in sub_coefs]
    coeffs_pt[i] = tuple(sub_coefs)

coeffs_pt = tuple(coeffs_pt)
background_pt = ptwt.waverec2(coeffs_pt, wavelet).squeeze()

background_pt_np = background_pt.numpy()