def idwt1(cA,cD,wname):
[lpd1,hpd1,lpr1,hpr1]=filter.filtcoef(wname)
len_lpfilt=int(len(lpr1))
len_hpfilt=int(len(hpr1))
len_avg=int(len_lpfilt/2 + len_hpfilt/2)
N= 2 * len(cD)
U=2
cA_up=sample.upsamp(cA,U)
cA_up=misc.per_ext(cA_up,int(len_avg/2))
X_lp=np.real(convol.convfft(cA_up,lpr1))
cD_up=sample.upsamp(cD,U)
cD_up=misc.per_ext(cD_up,int(len_avg/2))
X_hp=np.real(convol.convfft(cD_up,hpr1))
X_lp=X_lp[0:N+len_avg-1]
X_lp=X_lp[len_avg-1:]
X_hp=X_hp[0:N+len_avg-1]
X_hp=X_hp[len_avg-1:]
X=X_lp+X_hp
return X
def idwt1(cA, cD, wname):
[lpd1, hpd1, lpr1, hpr1] = filter.filtcoef(wname)
len_lpfilt = int(len(lpr1))
len_hpfilt = int(len(hpr1))
len_avg = int(len_lpfilt / 2 + len_hpfilt / 2)
N = 2 * len(cD)
U = 2
cA_up = sample.upsamp(cA, U)
cA_up = misc.per_ext(cA_up, int(len_avg / 2))
X_lp = np.real(convol.convfft(cA_up, lpr1))
cD_up = sample.upsamp(cD, U)
cD_up = misc.per_ext(cD_up, int(len_avg / 2))
X_hp = np.real(convol.convfft(cD_up, hpr1))
X_lp = X_lp[0 : N + len_avg - 1]
X_lp = X_lp[len_avg - 1 :]
X_hp = X_hp[0 : N + len_avg - 1]
X_hp = X_hp[len_avg - 1 :]
X = X_lp + X_hp
return X
def swt(sig, J, nm):
swtop = np.array([])
N = int(len(sig))
length = N
[lpd, hpd, lpr, hpr] = filter.filtcoef(nm)
for iter in range(J):
if iter > 0:
M = int(2 ** iter)
low_pass = sample.upsamp(lpd, M)
high_pass = sample.upsamp(hpd, M)
else:
low_pass = lpd
high_pass = hpd
len_filt = int(len(low_pass))
sig = misc.per_ext(sig, int(len_filt / 2))
cA = np.real(convol.convfft(sig, low_pass))
cD = np.real(convol.convfft(sig, high_pass))
cA = cA[len_filt:]
cA = cA[0:N]
cD = cD[len_filt:]
cD = cD[0:N]
sig = cA
if iter == J - 1:
swtop = np.append(cD, swtop)
swtop = np.append(cA, swtop)
else:
swtop = np.append(cD, swtop)
return swtop, length
def idwt1_sym(cA,cD,wname):
[lpd1,hpd1,lpr1,hpr1]=filter.filtcoef(wname)
if len(cA) > len(cD):
cA=cA[0:len(cD)]
len_lpfilt=int(len(lpr1))
len_hpfilt=int(len(hpr1))
lf=len_lpfilt
N= 2 * len(cD)
U=2
cA_up=sample.upsamp(cA,U)
cA_up=cA_up[0:len(cA_up)-1]
X_lp=np.real(convol.convfft(cA_up,lpr1))
cD_up=sample.upsamp(cD,U)
cD_up=cD_up[0:len(cD_up)-1]
X_hp=np.real(convol.convfft(cD_up,hpr1))
X=X_lp+X_hp
X=X[lf-2:]
X=X[:len(X)-lf+2]
return X
def swt(sig,J,nm):
swtop=np.array([])
N=int(len(sig))
length=N
[lpd,hpd,lpr,hpr]=filter.filtcoef(nm)
for iter in range(J):
if iter > 0:
M=int(2**iter)
low_pass=sample.upsamp(lpd,M)
high_pass=sample.upsamp(hpd,M)
else:
low_pass=lpd
high_pass=hpd
len_filt=int(len(low_pass))
sig=misc.per_ext(sig,int(len_filt/2))
cA=np.real(convol.convfft(sig,low_pass))
cD=np.real(convol.convfft(sig,high_pass))
cA=cA[len_filt:]
cA=cA[0:N]
cD=cD[len_filt:]
cD=cD[0:N]
sig=cA
if iter==J-1:
swtop=np.append(cD,swtop)
swtop=np.append(cA,swtop)
else:
swtop=np.append(cD,swtop)
return swtop,length
def idwt1_sym(cA, cD, wname):
[lpd1, hpd1, lpr1, hpr1] = filter.filtcoef(wname)
if len(cA) > len(cD):
cA = cA[0 : len(cD)]
len_lpfilt = int(len(lpr1))
len_hpfilt = int(len(hpr1))
lf = len_lpfilt
N = 2 * len(cD)
U = 2
cA_up = sample.upsamp(cA, U)
cA_up = cA_up[0 : len(cA_up) - 1]
X_lp = np.real(convol.convfft(cA_up, lpr1))
cD_up = sample.upsamp(cD, U)
cD_up = cD_up[0 : len(cD_up) - 1]
X_hp = np.real(convol.convfft(cD_up, hpr1))
X = X_lp + X_hp
X = X[lf - 2 :]
X = X[: len(X) - lf + 2]
return X
def swt2(inpsig,J,nm):
swtout=np.array([])
sig=np.array(inpsig)
m_size=int(np.size(sig,0))
n_size=int(np.size(sig,1))
rows_n=m_size
cols_n=n_size
[lp1,hp1,lp2,hp2]=filter.filtcoef(nm)
for iter in range(J):
U=int(2**iter)
low_pass=np.array([])
high_pass=np.array([])
if iter>0:
low_pass=sample.upsamp(lp1,U)
high_pass=sample.upsamp(hp1,U)
else:
low_pass=lp1
high_pass=hp1
lf=int(len(low_pass))
if int(np.size(sig,0)%2) == 0:
rows_n=int(np.size(sig,0))
else:
rows_n=int(np.size(sig,0)+1)
if int(np.size(sig,1)%2) == 0:
cols_n=int(np.size(sig,1))
else:
cols_n=int(np.size(sig,1)+1)
signal=np.ndarray(shape=(rows_n+lf,cols_n+lf))
signal=misc.per_ext2d(sig,lf/2)
len_x=int(np.size(signal,0))
len_y=int(np.size(signal,1))
sigL=np.ndarray(shape=(rows_n+lf,cols_n))
sigH=np.ndarray(shape=(rows_n+lf,cols_n))
cA=np.ndarray(shape=(rows_n,cols_n))
cH=np.ndarray(shape=(rows_n,cols_n))
cV=np.ndarray(shape=(rows_n,cols_n))
cD=np.ndarray(shape=(rows_n,cols_n))
for i in range(len_x):
temp_row=signal[i,0:len_y]
oup=np.real(convol.convfft(temp_row,low_pass))
oup=oup[lf:]
oup=oup[0:cols_n]
oup2=np.real(convol.convfft(temp_row,high_pass))
oup2=oup2[lf:]
oup2=oup2[0:cols_n]
sigL[i,:]=oup
sigH[i,:]=oup2
for j in range(cols_n):
temp_row=sigL[0:len_x,j]
oup=np.real(convol.convfft(temp_row,low_pass))
oup=oup[lf:]
oup=oup[0:rows_n]
oup2=np.real(convol.convfft(temp_row,high_pass))
oup2=oup2[lf:]
oup2=oup2[0:rows_n]
cA[:,j]=oup
cH[:,j]=oup2
for j in range(cols_n):
temp_row=sigH[0:len_x,j]
oup=np.real(convol.convfft(temp_row,low_pass))
oup=oup[lf:]
oup=oup[0:rows_n]
oup2=np.real(convol.convfft(temp_row,high_pass))
oup2=oup2[lf:]
oup2=oup2[0:rows_n]
cV[:,j]=oup
cD[:,j]=oup2
sig=cA
temp_sig2=np.array([])
if iter==J-1:
temp_sig2=np.reshape(cA,[np.size(cA,0)*np.size(cA,1)])
temp=np.reshape(cH,[np.size(cH,0)*np.size(cH,1)])
temp_sig2=np.concatenate([temp_sig2,temp])
temp=np.reshape(cV,[np.size(cV,0)*np.size(cV,1)])
temp_sig2=np.concatenate([temp_sig2,temp])
temp=np.reshape(cD,[np.size(cD,0)*np.size(cD,1)])
temp_sig2=np.concatenate([temp_sig2,temp])
swtout=np.concatenate([temp_sig2,swtout])
length=np.array([rows_n,cols_n])
return swtout,length
def iswt(swtop,J,nm):
N=int(len(swtop)/(J+1))
[lpd,hpd,lpr,hpr]=filter.filtcoef(nm)
low_pass=lpr
high_pass=hpr
lf=int(len(low_pass))
for iter in range(J):
iswt_output=np.zeros(N)
if iter==0:
appx_sig=swtop[0:N]
det_sig=swtop[N:2*N]
else:
det_sig=swtop[(iter+1)*N:(iter+2)*N]
value=int(2**(J-1-iter))
for count in range(value):
appx1=appx_sig[count:N:value]
det1=det_sig[count:N:value]
len1=len(appx1)
appx2=appx1[0:len1:2]
det2=det1[0:len1:2]
U=int(2)
cL0=sample.upsamp(appx2,U)
cH0=sample.upsamp(det2,U)
cL0=misc.per_ext(cL0,int(lf/2))
cH0=misc.per_ext(cH0,int(lf/2))
oup00L=np.real(convol.convfft(cL0,low_pass))
oup00H=np.real(convol.convfft(cH0,high_pass))
oup00L=oup00L[lf-1:]
oup00L=oup00L[0:len1]
oup00H=oup00H[lf-1:]
oup00H=oup00H[0:len1]
oup00=oup00L+oup00H
appx3=appx1[1:len1:2]
det3=det1[1:len1:2]
cL1=sample.upsamp(appx3,U)
cH1=sample.upsamp(det3,U)
cL1=misc.per_ext(cL1,int(lf/2))
cH1=misc.per_ext(cH1,int(lf/2))
oup01L=np.real(convol.convfft(cL1,low_pass))
oup01H=np.real(convol.convfft(cH1,high_pass))
oup01L=oup01L[lf-1:]
oup01L=oup01L[0:len1]
oup01H=oup01H[lf-1:]
oup01H=oup01H[0:len1]
oup01=oup01L+oup01H
oup01=misc.circshift(oup01,-1)
index2=int(0)
for index in xrange(count,N,value):
temp=(oup00[index2]+oup01[index2])*1.0/2.0
iswt_output[index]=temp
index2+=1
appx_sig=iswt_output
return iswt_output
def swt2(inpsig, J, nm):
swtout = np.array([])
sig = np.array(inpsig)
m_size = int(np.size(sig, 0))
n_size = int(np.size(sig, 1))
rows_n = m_size
cols_n = n_size
[lp1, hp1, lp2, hp2] = filter.filtcoef(nm)
for iter in range(J):
U = int(2 ** iter)
low_pass = np.array([])
high_pass = np.array([])
if iter > 0:
low_pass = sample.upsamp(lp1, U)
high_pass = sample.upsamp(hp1, U)
else:
low_pass = lp1
high_pass = hp1
lf = int(len(low_pass))
if int(np.size(sig, 0) % 2) == 0:
rows_n = int(np.size(sig, 0))
else:
rows_n = int(np.size(sig, 0) + 1)
if int(np.size(sig, 1) % 2) == 0:
cols_n = int(np.size(sig, 1))
else:
cols_n = int(np.size(sig, 1) + 1)
signal = np.ndarray(shape=(rows_n + lf, cols_n + lf))
signal = misc.per_ext2d(sig, lf / 2)
len_x = int(np.size(signal, 0))
len_y = int(np.size(signal, 1))
sigL = np.ndarray(shape=(rows_n + lf, cols_n))
sigH = np.ndarray(shape=(rows_n + lf, cols_n))
cA = np.ndarray(shape=(rows_n, cols_n))
cH = np.ndarray(shape=(rows_n, cols_n))
cV = np.ndarray(shape=(rows_n, cols_n))
cD = np.ndarray(shape=(rows_n, cols_n))
for i in range(len_x):
temp_row = signal[i, 0:len_y]
oup = np.real(convol.convfft(temp_row, low_pass))
oup = oup[lf:]
oup = oup[0:cols_n]
oup2 = np.real(convol.convfft(temp_row, high_pass))
oup2 = oup2[lf:]
oup2 = oup2[0:cols_n]
sigL[i, :] = oup
sigH[i, :] = oup2
for j in range(cols_n):
temp_row = sigL[0:len_x, j]
oup = np.real(convol.convfft(temp_row, low_pass))
oup = oup[lf:]
oup = oup[0:rows_n]
oup2 = np.real(convol.convfft(temp_row, high_pass))
oup2 = oup2[lf:]
oup2 = oup2[0:rows_n]
cA[:, j] = oup
cH[:, j] = oup2
for j in range(cols_n):
temp_row = sigH[0:len_x, j]
oup = np.real(convol.convfft(temp_row, low_pass))
oup = oup[lf:]
oup = oup[0:rows_n]
oup2 = np.real(convol.convfft(temp_row, high_pass))
oup2 = oup2[lf:]
oup2 = oup2[0:rows_n]
cV[:, j] = oup
cD[:, j] = oup2
sig = cA
temp_sig2 = np.array([])
if iter == J - 1:
temp_sig2 = np.reshape(cA, [np.size(cA, 0) * np.size(cA, 1)])
temp = np.reshape(cH, [np.size(cH, 0) * np.size(cH, 1)])
temp_sig2 = np.concatenate([temp_sig2, temp])
temp = np.reshape(cV, [np.size(cV, 0) * np.size(cV, 1)])
temp_sig2 = np.concatenate([temp_sig2, temp])
temp = np.reshape(cD, [np.size(cD, 0) * np.size(cD, 1)])
temp_sig2 = np.concatenate([temp_sig2, temp])
swtout = np.concatenate([temp_sig2, swtout])
length = np.array([rows_n, cols_n])
return swtout, length
def iswt(swtop, J, nm):
N = int(len(swtop) / (J + 1))
[lpd, hpd, lpr, hpr] = filter.filtcoef(nm)
low_pass = lpr
high_pass = hpr
lf = int(len(low_pass))
for iter in range(J):
iswt_output = np.zeros(N)
if iter == 0:
appx_sig = swtop[0:N]
det_sig = swtop[N : 2 * N]
else:
det_sig = swtop[(iter + 1) * N : (iter + 2) * N]
value = int(2 ** (J - 1 - iter))
for count in range(value):
appx1 = appx_sig[count:N:value]
det1 = det_sig[count:N:value]
len1 = len(appx1)
appx2 = appx1[0:len1:2]
det2 = det1[0:len1:2]
U = int(2)
cL0 = sample.upsamp(appx2, U)
cH0 = sample.upsamp(det2, U)
cL0 = misc.per_ext(cL0, int(lf / 2))
cH0 = misc.per_ext(cH0, int(lf / 2))
oup00L = np.real(convol.convfft(cL0, low_pass))
oup00H = np.real(convol.convfft(cH0, high_pass))
oup00L = oup00L[lf - 1 :]
oup00L = oup00L[0:len1]
oup00H = oup00H[lf - 1 :]
oup00H = oup00H[0:len1]
oup00 = oup00L + oup00H
appx3 = appx1[1:len1:2]
det3 = det1[1:len1:2]
cL1 = sample.upsamp(appx3, U)
cH1 = sample.upsamp(det3, U)
cL1 = misc.per_ext(cL1, int(lf / 2))
cH1 = misc.per_ext(cH1, int(lf / 2))
oup01L = np.real(convol.convfft(cL1, low_pass))
oup01H = np.real(convol.convfft(cH1, high_pass))
oup01L = oup01L[lf - 1 :]
oup01L = oup01L[0:len1]
oup01H = oup01H[lf - 1 :]
oup01H = oup01H[0:len1]
oup01 = oup01L + oup01H
oup01 = misc.circshift(oup01, -1)
index2 = int(0)
for index in xrange(count, N, value):
temp = (oup00[index2] + oup01[index2]) * 1.0 / 2.0
iswt_output[index] = temp
index2 += 1
appx_sig = iswt_output
return iswt_output
