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 dwt1(signal,wname):
[lpd,hpd,lpr,hpr]=filter.filtcoef(wname)
len_lpfilt=int(len(lpd))
len_hpfilt=int(len(hpd))
len_avg=int(len_lpfilt/2 + len_hpfilt/2)
len_sig = int( 2 *(np.ceil(len(signal) * 1.0/2.0)))
D=int(2)
signal=misc.per_ext(signal,int(len_avg/2))
cA_undec=np.real(convol.convfft(signal,lpd))
cA_undec=cA_undec[len_avg-1:]
cA_undec=cA_undec[:len(cA_undec)-len_avg+1]
cA_undec=cA_undec[1:len_sig]
cA=sample.downsamp(cA_undec,D)
cD_undec=np.real(convol.convfft(signal,hpd))
cD_undec=cD_undec[len_avg-1:]
cD_undec=cD_undec[:len(cD_undec)-len_avg+1]
cD_undec=cD_undec[1:len_sig]
cD=sample.downsamp(cD_undec,D)
return cA,cD
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 dwt2(signal,J,nm,ext):
flag=np.array([])
dwtout=np.array([])
length=np.array([])
sig=np.array(signal)
rows_n=int(np.size(sig,0))
cols_n=int(np.size(sig,1))
Max_Iter=min(int(np.ceil(np.log2(rows_n))),int(np.ceil(np.log2(cols_n))))
if Max_Iter < J:
print J," Iterations are not possible with signals of this dimension "
flag=np.append(flag,J)
flag=np.append(flag,0)
if ext == 'per':
flag=np.append(flag,int(0))
else:
flag=np.append(flag,int(1))
length=np.append(int(cols_n),length)
length=np.append(int(rows_n),length)
orig=sig;
[lp1,hp1,lp2,hp2]=filter.filtcoef(nm)
lf=len(lp1)
for iter in range(J):
if ext=='per':
rows_n=int(np.ceil(rows_n*1.0/2.0))
cols_n=int(np.ceil(cols_n*1.0/2.0))
else:
rows_n=int(np.floor((rows_n+lf-1)*1.0/2.0))
cols_n=int(np.floor((cols_n+lf-1)*1.0/2.0))
length=np.append(int(cols_n),length)
length=np.append(int(rows_n),length)
if ext=='per':
[cA,cH,cV,cD]=dwt2_per(orig,nm)
else:
[cA,cH,cV,cD]=dwt2_sym(orig,nm)
temp_sig2=np.array([])
orig=cA
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])
dwtout=np.concatenate([temp_sig2,dwtout])
return dwtout,length,flag
def dwt2(signal, J, nm, ext):
flag = np.array([])
dwtout = np.array([])
length = np.array([])
sig = np.array(signal)
rows_n = int(np.size(sig, 0))
cols_n = int(np.size(sig, 1))
Max_Iter = min(int(np.ceil(np.log2(rows_n))), int(np.ceil(np.log2(cols_n))))
if Max_Iter < J:
print J, " Iterations are not possible with signals of this dimension "
flag = np.append(flag, J)
flag = np.append(flag, 0)
if ext == "per":
flag = np.append(flag, int(0))
else:
flag = np.append(flag, int(1))
length = np.append(int(cols_n), length)
length = np.append(int(rows_n), length)
orig = sig
[lp1, hp1, lp2, hp2] = filter.filtcoef(nm)
lf = len(lp1)
for iter in range(J):
if ext == "per":
rows_n = int(np.ceil(rows_n * 1.0 / 2.0))
cols_n = int(np.ceil(cols_n * 1.0 / 2.0))
else:
rows_n = int(np.floor((rows_n + lf - 1) * 1.0 / 2.0))
cols_n = int(np.floor((cols_n + lf - 1) * 1.0 / 2.0))
length = np.append(int(cols_n), length)
length = np.append(int(rows_n), length)
if ext == "per":
[cA, cH, cV, cD] = dwt2_per(orig, nm)
else:
[cA, cH, cV, cD] = dwt2_sym(orig, nm)
temp_sig2 = np.array([])
orig = cA
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])
dwtout = np.concatenate([temp_sig2, dwtout])
return dwtout, length, flag
def dwt1_sym(signal, wname):
[lpd, hpd, lpr, hpr] = filter.filtcoef(wname)
len_lpfilt = int(len(lpd))
len_hpfilt = int(len(hpd))
lf = len_lpfilt
D = int(2)
signal = misc.symm_ext(signal, int(lf - 1))
cA_undec = np.real(convol.convfft(signal, lpd))
cA_undec = cA_undec[lf:]
cA_undec = cA_undec[: len(cA_undec) - lf + 1]
cA = sample.downsamp(cA_undec, D)
cD_undec = np.real(convol.convfft(signal, hpd))
cD_undec = cD_undec[lf:]
cD_undec = cD_undec[: len(cD_undec) - lf + 1]
cD = sample.downsamp(cD_undec, D)
return cA, cD
def dwt1_sym(signal,wname):
[lpd,hpd,lpr,hpr]=filter.filtcoef(wname)
len_lpfilt=int(len(lpd))
len_hpfilt=int(len(hpd))
lf=len_lpfilt
D=int(2)
signal=misc.symm_ext(signal,int(lf - 1))
cA_undec=np.real(convol.convfft(signal,lpd))
cA_undec=cA_undec[lf:]
cA_undec=cA_undec[:len(cA_undec)-lf+1]
cA=sample.downsamp(cA_undec,D)
cD_undec=np.real(convol.convfft(signal,hpd))
cD_undec=cD_undec[lf:]
cD_undec=cD_undec[:len(cD_undec)-lf+1]
cD=sample.downsamp(cD_undec,D)
return cA,cD
def dwt1(signal, wname):
[lpd, hpd, lpr, hpr] = filter.filtcoef(wname)
len_lpfilt = int(len(lpd))
len_hpfilt = int(len(hpd))
len_avg = int(len_lpfilt / 2 + len_hpfilt / 2)
len_sig = int(2 * (np.ceil(len(signal) * 1.0 / 2.0)))
D = int(2)
signal = misc.per_ext(signal, int(len_avg / 2))
cA_undec = np.real(convol.convfft(signal, lpd))
cA_undec = cA_undec[len_avg - 1 :]
cA_undec = cA_undec[: len(cA_undec) - len_avg + 1]
cA_undec = cA_undec[1:len_sig]
cA = sample.downsamp(cA_undec, D)
cD_undec = np.real(convol.convfft(signal, hpd))
cD_undec = cD_undec[len_avg - 1 :]
cD_undec = cD_undec[: len(cD_undec) - len_avg + 1]
cD_undec = cD_undec[1:len_sig]
cD = sample.downsamp(cD_undec, D)
return cA, cD
def dwt2_sym(signal,name):
rows=int(np.size(signal,0))
cols=int(np.size(signal,1))
[lp1,hp1,lp2,hp2]=filter.filtcoef(name)
lf=len(lp1)
rows_n=int(np.floor((rows+lf-1)*1.0/2.0))
cols_n=int(np.floor((cols+lf-1)*1.0/2.0))
lp_dn1=np.ndarray(shape=(rows,cols_n))
hp_dn1=np.ndarray(shape=(rows,cols_n))
cLL=np.ndarray(shape=(rows_n,cols_n))
cLH=np.ndarray(shape=(rows_n,cols_n))
cHL=np.ndarray(shape=(rows_n,cols_n))
cHH=np.ndarray(shape=(rows_n,cols_n))
for i in range(rows):
temp_row=signal[i,:]
[oup_lp,oup_hp]=dwt1_sym(temp_row,name)
lp_dn1[i,:]=oup_lp
hp_dn1[i,:]=oup_hp
cols=cols_n
for j in range(cols):
temp_row3=lp_dn1[:,j]
[oup_lp,oup_hp]=dwt1_sym(temp_row3,name)
cLL[:,j]=oup_lp
cLH[:,j]=oup_hp
for j in range(cols):
temp_row5=hp_dn1[:,j]
[oup_lp,oup_hp]=dwt1_sym(temp_row5,name)
cHL[:,j]=oup_lp
cHH[:,j]=oup_hp
return cLL,cLH,cHL,cHH
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 dwt2_per(signal, name):
rows = int(np.size(signal, 0))
cols = int(np.size(signal, 1))
rows_n = int(np.ceil(rows * 1.0 / 2.0))
cols_n = int(np.ceil(cols * 1.0 / 2.0))
lp_dn1 = np.ndarray(shape=(rows, cols_n))
hp_dn1 = np.ndarray(shape=(rows, cols_n))
cLL = np.ndarray(shape=(rows_n, cols_n))
cLH = np.ndarray(shape=(rows_n, cols_n))
cHL = np.ndarray(shape=(rows_n, cols_n))
cHH = np.ndarray(shape=(rows_n, cols_n))
[lp1, hp1, lp2, hp2] = filter.filtcoef(name)
for i in range(rows):
temp_row = signal[i, :]
[oup_lp, oup_hp] = dwt1(temp_row, name)
lp_dn1[i, :] = oup_lp
hp_dn1[i, :] = oup_hp
cols = cols_n
for j in range(cols):
temp_row3 = lp_dn1[:, j]
[oup_lp, oup_hp] = dwt1(temp_row3, name)
cLL[:, j] = oup_lp
cLH[:, j] = oup_hp
for j in range(cols):
temp_row5 = hp_dn1[:, j]
[oup_lp, oup_hp] = dwt1(temp_row5, name)
cHL[:, j] = oup_lp
cHH[:, j] = oup_hp
return cLL, cLH, cHL, cHH
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 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 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 idwt2(dwtop,nm,length,flag):
J=int(flag[0])
rows=int(length[0])
cols=int(length[1])
sum_coef=int(0)
[lp1,hp1,lp2,hp2]=filter.filtcoef(nm)
lf=len(lp1)
for iter in range(J):
rows_n=int(length[2*int(iter)])
cols_n=int(length[2*int(iter)+1])
if iter==0:
temp=dwtop[0:cols_n*rows_n]
cLL=np.reshape(temp,[rows_n,cols_n])
temp=dwtop[cols_n*rows_n:2*cols_n*rows_n]
cLH=np.reshape(temp,[rows_n,cols_n])
temp=dwtop[2*cols_n*rows_n:3*cols_n*rows_n]
cHL=np.reshape(temp,[rows_n,cols_n])
temp=dwtop[3*cols_n*rows_n:4*cols_n*rows_n]
cHH=np.reshape(temp,[rows_n,cols_n])
else:
temp=dwtop[sum_coef:sum_coef+rows_n*cols_n]
cLH=np.reshape(temp,[rows_n,cols_n])
temp=dwtop[sum_coef+rows_n*cols_n:sum_coef+2*rows_n*cols_n]
cHL=np.reshape(temp,[rows_n,cols_n])
temp=dwtop[sum_coef+2*rows_n*cols_n:sum_coef+3*rows_n*cols_n]
cHH=np.reshape(temp,[rows_n,cols_n])
len_x=np.size(cLH,0)
len_y=np.size(cLH,1)
if flag[2]==0:
cL=np.ndarray(shape=(2*len_x,len_y))
cH=np.ndarray(shape=(2*len_x,len_y))
t_iter=2*len_x
else:
cL=np.ndarray(shape=(2*len_x-lf+2,len_y))
cH=np.ndarray(shape=(2*len_x-lf+2,len_y))
t_iter=2*len_x-lf+2
if iter==0:
for j in range(len_y):
sigLL=cLL[0:len_x,j]
sigLH=cLH[0:len_x,j]
if int(flag[2])==0:
oup=idwt1(sigLL,sigLH,nm)
cL[:,j]=oup
else:
oup=idwt1_sym(sigLL,sigLH,nm)
cL[:,j]=oup
else:
rows1=int(np.size(cLH,0))
cols1=int(np.size(cLH,1))
for j in range(cols1):
sigLL=cLL[0:rows1,j]
sigLH=cLH[0:rows1,j]
if int(flag[2])==0:
oup=idwt1(sigLL,sigLH,nm)
cL[:,j]=oup
else:
oup=idwt1_sym(sigLL,sigLH,nm)
cL[:,j]=oup
for j in range(len_y):
sigHL=cHL[0:len_x,j]
sigHH=cHH[0:len_x,j]
if int(flag[2])==0:
oup=idwt1(sigHL,sigHH,nm)
cH[:,j]=oup
else:
oup=idwt1_sym(sigHL,sigHH,nm)
cH[:,j]=oup
if int(flag[2])==0:
signal=np.ndarray(shape=(2*len_x,2*len_y))
else:
signal=np.ndarray(shape=(2*len_x-lf+2,2*len_y-lf+2))
for i in range(t_iter):
sigL=cL[i,0:len_y]
sigH=cH[i,0:len_y]
if int(flag[2])==0:
oup=idwt1(sigL,sigH,nm)
signal[i,:]=oup
else:
oup=idwt1_sym(sigL,sigH,nm)
signal[i,:]=oup
idwt_output=signal
if iter==0:
sum_coef+=4*rows_n*cols_n
else:
sum_coef+=3*rows_n*cols_n
cLL=signal
len_length=int(len(length))
idwt_output=idwt_output[0:int(length[len_length-2]),0:int(length[len_length-1])]
return idwt_output
def idwt2(dwtop, nm, length, flag):
J = int(flag[0])
rows = int(length[0])
cols = int(length[1])
sum_coef = int(0)
[lp1, hp1, lp2, hp2] = filter.filtcoef(nm)
lf = len(lp1)
for iter in range(J):
rows_n = int(length[2 * int(iter)])
cols_n = int(length[2 * int(iter) + 1])
if iter == 0:
temp = dwtop[0 : cols_n * rows_n]
cLL = np.reshape(temp, [rows_n, cols_n])
temp = dwtop[cols_n * rows_n : 2 * cols_n * rows_n]
cLH = np.reshape(temp, [rows_n, cols_n])
temp = dwtop[2 * cols_n * rows_n : 3 * cols_n * rows_n]
cHL = np.reshape(temp, [rows_n, cols_n])
temp = dwtop[3 * cols_n * rows_n : 4 * cols_n * rows_n]
cHH = np.reshape(temp, [rows_n, cols_n])
else:
temp = dwtop[sum_coef : sum_coef + rows_n * cols_n]
cLH = np.reshape(temp, [rows_n, cols_n])
temp = dwtop[sum_coef + rows_n * cols_n : sum_coef + 2 * rows_n * cols_n]
cHL = np.reshape(temp, [rows_n, cols_n])
temp = dwtop[sum_coef + 2 * rows_n * cols_n : sum_coef + 3 * rows_n * cols_n]
cHH = np.reshape(temp, [rows_n, cols_n])
len_x = np.size(cLH, 0)
len_y = np.size(cLH, 1)
if flag[2] == 0:
cL = np.ndarray(shape=(2 * len_x, len_y))
cH = np.ndarray(shape=(2 * len_x, len_y))
t_iter = 2 * len_x
else:
cL = np.ndarray(shape=(2 * len_x - lf + 2, len_y))
cH = np.ndarray(shape=(2 * len_x - lf + 2, len_y))
t_iter = 2 * len_x - lf + 2
if iter == 0:
for j in range(len_y):
sigLL = cLL[0:len_x, j]
sigLH = cLH[0:len_x, j]
if int(flag[2]) == 0:
oup = idwt1(sigLL, sigLH, nm)
cL[:, j] = oup
else:
oup = idwt1_sym(sigLL, sigLH, nm)
cL[:, j] = oup
else:
rows1 = int(np.size(cLH, 0))
cols1 = int(np.size(cLH, 1))
for j in range(cols1):
sigLL = cLL[0:rows1, j]
sigLH = cLH[0:rows1, j]
if int(flag[2]) == 0:
oup = idwt1(sigLL, sigLH, nm)
cL[:, j] = oup
else:
oup = idwt1_sym(sigLL, sigLH, nm)
cL[:, j] = oup
for j in range(len_y):
sigHL = cHL[0:len_x, j]
sigHH = cHH[0:len_x, j]
if int(flag[2]) == 0:
oup = idwt1(sigHL, sigHH, nm)
cH[:, j] = oup
else:
oup = idwt1_sym(sigHL, sigHH, nm)
cH[:, j] = oup
if int(flag[2]) == 0:
signal = np.ndarray(shape=(2 * len_x, 2 * len_y))
else:
signal = np.ndarray(shape=(2 * len_x - lf + 2, 2 * len_y - lf + 2))
for i in range(t_iter):
sigL = cL[i, 0:len_y]
sigH = cH[i, 0:len_y]
if int(flag[2]) == 0:
oup = idwt1(sigL, sigH, nm)
signal[i, :] = oup
else:
oup = idwt1_sym(sigL, sigH, nm)
signal[i, :] = oup
idwt_output = signal
if iter == 0:
sum_coef += 4 * rows_n * cols_n
else:
sum_coef += 3 * rows_n * cols_n
cLL = signal
len_length = int(len(length))
idwt_output = idwt_output[0 : int(length[len_length - 2]), 0 : int(length[len_length - 1])]
return idwt_output