def generate_prediction(AL, BL, CL, AH, CH):
flow_AL_BL = motion_estimation(AL, BL)
flow_CL_BL = motion_estimation(CL, BL)
BAH = estimate_frame(AH, flow_AL_BL)
BCH = estimate_frame(CH, flow_CL_BL)
BAL = estimate_frame(AL, flow_AL_BL)
BCL = estimate_frame(CL, flow_CL_BL)
EAL = BL - BAL
ECL = BL - BCL
SAL = 1 / (1 + abs(EAL))
SCL = 1 / (1 + abs(ECL))
return (BAH * SAL + BCH * SCL) / (SAL + SCL)
def __backward_butterfly(self, aL, aH, bL, residue_bH, cL, cH):
'''Backward MCDWT butterfly.
Input:
-----
aL, aH, bL, residue_bH, cL, cH: array[y, x, component], the
\tilde{b} image.
Output:
------
bH: array[y, x, component], the high-frequencies of the image
b.
'''
#AL = self.dwt.backward((aL, self.zero_H))
#BL = self.dwt.backward((bL, self.zero_H))
#CL = self.dwt.backward((cL, self.zero_H))
AH = self.dwt.backward((self.zero_L, aH))
residue_BH = self.dwt.backward((self.zero_L, residue_bH))
CH = self.dwt.backward((self.zero_L, cH))
#prediction_BH = predictor.generate_prediction(AL, BL, CL, AH, CH)
#prediction_BH = predictor.generate_rediction(aL, bL, cL, aH, cH)
flow_aL_bL = motion_estimation(aL, bL)
flow_cL_bL = motion_estimation(cL, bL)
c1 = pywt.idwt2((flow_aL_bL[:, :, 0], (None, None, None)),
wavelet='haar',
mode='per')
c2 = pywt.idwt2((flow_aL_bL[:, :, 0], (None, None, None)),
wavelet='haar',
mode='per')
Flow_aL_bL = np.stack((c1, c2), axis=-1)
c1 = pywt.idwt2((flow_cL_bL[:, :, 0], (None, None, None)),
wavelet='haar',
mode='per')
c2 = pywt.idwt2((flow_cL_bL[:, :, 0], (None, None, None)),
wavelet='haar',
mode='per')
Flow_cL_bL = np.stack((c1, c2), axis=-1)
BAH = estimate_frame(AH, Flow_aL_bL)
BCH = estimate_frame(CH, Flow_cL_bL)
prediction_BH = (BAH + BCH) / 2
BH = residue_BH + prediction_BH
bH = self.dwt.forward(BH)
return bH[1]
def __forward_butterfly(self, aL, aH, bL, bH, cL, cH):
'''Forward MCDWT butterfly.
Input:
-----
aL, aH, bL, bH, cL, cH: array[y, x, component], the decomposition of
the images a, b and c.
Output:
------
residue_bH: array[y, x, component], the residue of the
high-requency information of the image b.
'''
#AL = self.dwt.backward((aL, self.zero_H))
#BL = self.dwt.backward((bL, self.zero_H))
#CL = self.dwt.backward((cL, self.zero_H))
AH = self.dwt.backward((self.zero_L, aH))
BH = self.dwt.backward((self.zero_L, bH))
CH = self.dwt.backward((self.zero_L, cH))
#prediction_BH = predictor.generate_prediction(AL, BL, CL, AH, CH)
#prediction_BH = predictor.generate_prediction(aL, bL, cL, aH, cH)
flow_aL_bL = motion_estimation(aL, bL)
flow_cL_bL = motion_estimation(cL, bL)
c1 = pywt.idwt2((flow_aL_bL[:, :, 0], (None, None, None)),
wavelet='haar',
mode='per')
c2 = pywt.idwt2((flow_aL_bL[:, :, 0], (None, None, None)),
wavelet='haar',
mode='per')
Flow_aL_bL = np.stack((c1, c2), axis=-1)
c1 = pywt.idwt2((flow_cL_bL[:, :, 0], (None, None, None)),
wavelet='haar',
mode='per')
c2 = pywt.idwt2((flow_cL_bL[:, :, 0], (None, None, None)),
wavelet='haar',
mode='per')
Flow_cL_bL = np.stack((c1, c2), axis=-1)
BAH = estimate_frame(AH, Flow_aL_bL)
BCH = estimate_frame(CH, Flow_cL_bL)
prediction_BH = (BAH + BCH) / 2
residue_BH = BH - prediction_BH
residue_bH = self.dwt.forward(residue_BH)
return residue_bH[1]
def __backward_butterfly(self, aL, aH, bL, residue_bH, cL, cH):
AL = self.dwt.backward((aL, self.zero_H))
BL = self.dwt.backward((bL, self.zero_H))
CL = self.dwt.backward((cL, self.zero_H))
AH = self.dwt.backward((self.zero_L, aH))
residue_BH = self.dwt.backward((self.zero_L, residue_bH))
CH = self.dwt.backward((self.zero_L, cH))
'''
Cálculo BHA, BLA, BHC, BLC optimizado
'''
flow = motion_estimation(AL, BL)
BHA = estimate_frame(AH, flow)
BLA = estimate_frame(AL, flow)
flow = motion_estimation(CL, BL)
BHC = estimate_frame(CH, flow)
BLC = estimate_frame(CL, flow)
prediction_BH = defprediction.calcula_prediction()
BH = residue_BH + prediction_BH
bH = self.dwt.forward(BH)
return bH[1]
def __forward_butterfly(self, aL, aH, bL, bH, cL, cH):
'''Motion compensated forward MCDWT butterfly.
Input:
-----
aL, aH, bL, bH, cL, cH: array[y, x, component], the decomposition of
the images a, b and c.
Output:
------
residue_bH: array[y, x, component], the base of the decomposition of
the residue fot the image b.
'''
AL = self.dwt.backward((aL, self.zero_H))
BL = self.dwt.backward((bL, self.zero_H))
CL = self.dwt.backward((cL, self.zero_H))
AH = self.dwt.backward((self.zero_L, aH))
BH = self.dwt.backward((self.zero_L, bH))
CH = self.dwt.backward((self.zero_L, cH))
'''
Cálculo BHA, BLA, BHC, BLC optimizado
'''
flow = motion_estimation(AL, BL)
BHA = estimate_frame(AH, flow)
BLA = estimate_frame(AL, flow)
flow = motion_estimation(CL, BL)
BHC = estimate_frame(CH, flow)
BLC = estimate_frame(CL, flow)
prediction_BH = defprediction.calcula_prediction()
residue_BH = BH - prediction_BH
residue_bH = self.dwt.forward(residue_BH)
return residue_bH[1]
def generate_prediction(AL, BL, CL, AH, CH):
flow_CL_BL = motion_estimation(CL, BL)
BCH = estimate_frame(CH, flow_CL_BL)
return BCH
def generate_prediction(AL, BL, CL, AH, CH):
flow_AL_BL = motion_estimation(AL, BL)
BAH = estimate_frame(AH, flow_AL_BL)
return BAH