def getFilterMatrix(imgList, const):
xlim,ylim = const.imgSize
shortToLongM = np.zeros((xlim*ylim,len(imgList)))
longToShortM = np.zeros((len(imgList),xlim*ylim))
for k in range(len(imgList)):
i, j, _, _ = imgList[k]
shortToLongM[i*ylim+j][k] = 1
longToShortM[k][i*ylim+j] = 1
return commons.sMatrix(shortToLongM), commons.sMatrix(longToShortM)
def getPSFMatrix(psfFileName,imgList,const):
psfMatrix, _= commons.readFitsImage(psfFileName)
normalizedPSF = psfMatrix/np.sum(psfMatrix)
M, N = const.imgSize
fullB = np.zeros((M*N, M*N))
xlim, ylim = normalizedPSF.shape
#B = np.zeros((const.length, const.length))
for u in range(xlim):
for v in range(ylim):
for h in range(M*N):
if h+u>=0 and h+u<=M-1 and h+v>=0 and h+v<=N-1:
g=(h+u)+(h+v-1)*(M-1)
fullB[g][h]=normalizedPSF[u][v]
fullB = commons.sMatrix(fullB)
shortTolongM, longToShortM = commons.getFilterMatrix(imgList, const)
B = longToShortM*fullB*shortTolongM
#print longToShortM.todense()
#print scipy.linalg.norm((longToShortM*fullB*shortTolongM).todense())
#print shortTolongM.todense()
return B
def getLensOperator(mappingDict, srcBrightNess):
dim = len(mappingDict)
L = np.zeros((dim, 2*dim))
imgPointList = mappingDict.keys()
srcPointList = mappingDict.values()
srcPosition = []
firstOrderWeightList = []
secondOrderWeightList = []
Hsy1 = np.zeros((dim, dim))
Hsy2 = np.zeros((dim, dim))
d = np.zeros(dim)
normV = [[] for i in range(dim)]
#srcBrightNess = np.ones(dim)
varList = []
RTR = np.zeros((2*dim, 2*dim))
for i in range(dim):
imgX, imgY = imgPointList[i]
srcX, srcY, imgBrightNess, type, var = srcPointList[i]
srcPosition.append((srcX, srcY))
varList.append(var)
d[i] = imgBrightNess
if type=='o':
if (imgX, imgY-1) in imgPointList and (imgX, imgY+1) in imgPointList and (imgX+1, imgY) in imgPointList:
Hsy1[i][i]=1
Hsy2[i][i]=1
w1Index = imgPointList.index((imgX, imgY-1))
w2Index = imgPointList.index((imgX, imgY+1))
w3Index = imgPointList.index((imgX+1, imgY))
A = (mappingDict[(imgX,imgY-1)][0], mappingDict[(imgX,imgY-1)][1])
B = (mappingDict[(imgX,imgY+1)][0], mappingDict[(imgX,imgY+1)][1])
C = (mappingDict[(imgX+1,imgY)][0], mappingDict[(imgX+1,imgY)][1])
L[i][w1Index], L[i][w2Index],L[i][w3Index] = commons.getTriWeight(A,B,C, (srcX, srcY))
# update the normV
An = (A[0], A[1], srcBrightNess[w1Index])
Bn = (B[0], B[1], srcBrightNess[w2Index])
Cn = (C[0], C[1], srcBrightNess[w3Index])
n0, n1, n2 = commons.getNormVectors(An, Bn, Cn)
normV[w1Index].append((n0, n1, n2))
normV[w2Index].append((n0, n1, n2))
normV[w3Index].append((n0, n1, n2))
firstOrderWeightList.append((w1Index, w2Index, w3Index, L[i][w1Index], L[i][w2Index],L[i][w3Index]))
else:
L[i][i] =1
firstOrderWeightList.append((i, i, i , 1/3.,1/3.,1/3.))
if type=='v':
L[i][i] =1
firstOrderWeightList.append((i, i, i , 1/3.,1/3.,1/3.))
if (imgX-1, imgY-1) in imgPointList and (imgX-1, imgY+1) in imgPointList and (imgX+1, imgY-1) in imgPointList and (imgX+1, imgY+1) in imgPointList:
wAIndex = imgPointList.index((imgX-1, imgY-1))
wBIndex = imgPointList.index((imgX-1, imgY+1))
wCIndex = imgPointList.index((imgX+0, imgY+0))
wDIndex = imgPointList.index((imgX+1, imgY-1))
wEIndex = imgPointList.index((imgX+1, imgY+1))
A = (mappingDict[(imgX-1,imgY-1)][0], mappingDict[(imgX-1,imgY-1)][1])
B = (mappingDict[(imgX-1,imgY+1)][0], mappingDict[(imgX-1,imgY+1)][1])
C = (mappingDict[(imgX+0,imgY+0)][0], mappingDict[(imgX+0,imgY+0)][1])
D = (mappingDict[(imgX+1,imgY-1)][0], mappingDict[(imgX+1,imgY-1)][1])
E = (mappingDict[(imgX+1,imgY+1)][0], mappingDict[(imgX+1,imgY+1)][1])
Hsy1[i][wAIndex], Hsy1[i][wBIndex], Hsy1[i][wCIndex], Hsy1[i][wDIndex],Hsy1[i][wEIndex],\
Hsy2[i][wAIndex], Hsy2[i][wBIndex], Hsy2[i][wCIndex], Hsy2[i][wDIndex],Hsy2[i][wEIndex] = commons.getPentWeigth(A, B, C, D, E)
HsTHs = scipy.sparse.coo_matrix(Hsy1.transpose())*scipy.sparse.coo_matrix(Hsy1) + \
scipy.sparse.coo_matrix(Hsy2.transpose())*scipy.sparse.coo_matrix(Hsy2)
HsTHs = HsTHs.toarray()
#Hs = np.linalg.cholesky(tt)
for i in range(dim):
for j in range(dim):
RTR[i][j] = HsTHs[i][j]
# Diagonal covariance matrix:
C = commons.listToDiagonalMatrix(varList)
normV = commons.getMeanNorm(normV)
return srcPosition, srcPointList, commons.sMatrix(L), normV, C, firstOrderWeightList, d, RTR, HsTHs, imgPointList