def goRangeFunctions():
xmin, xmax, sigma = -3.5, 3.5, 1.0
nx = 351
dx = (xmax - xmin) / (nx - 1)
fx = xmin
sx = Sampling(nx, dx, fx)
yg = BilateralFilter.sampleRangeFunction(gauss, sigma, sx)
yt = BilateralFilter.sampleRangeFunction(tukey, sigma, sx)
#yh = BilateralFilter.sampleRangeFunction(huber,sigma,sx)
#x = rampfloat(fx,dx,nx)
#yg = mul(x,yg)
#yh = mul(x,yh)
#yt = mul(x,yt)
sp = SimplePlot()
sp.setBackground(backgroundColor)
sp.setFontSizeForSlide(1.0, 0.9)
sp.setSize(700, 500)
sp.setHLabel(" ")
sp.setVLabel(" ")
solid = PointsView.Line.SOLID
dash = PointsView.Line.DASH
dot = PointsView.Line.DOT
#pv = sp.addPoints(sx,yh); pv.setLineStyle(dot); pv.setLineWidth(4)
pv = sp.addPoints(sx, yg)
pv.setLineStyle(dash)
pv.setLineWidth(4)
pv = sp.addPoints(sx, yt)
pv.setLineStyle(solid)
pv.setLineWidth(4)
if pngDir:
sp.paintToPng(720, 3.3, pngDir + "blftg.png")
def goBilateral(f):
g = like(f)
sigmaS = 24.0 # spatial filter sigma
sigmaR = computeSigmaR(f)
blf = BilateralFilter(sigmaS, sigmaR)
blf.apply(f, g)
plot(g, cmin=-1, cmax=1, name='bilateral filter')
def goBilateral(f): g = like(f) sigmaS = 24.0 # spatial filter sigma sigmaR = computeSigmaR(f) blf = BilateralFilter(sigmaS,sigmaR) blf.apply(f,g) plot(g,cmin=-1,cmax=1,name='bilateral filter')
def goRangeFunctions():
xmin,xmax,sigma = -3.5,3.5,1.0
nx = 351
dx = (xmax-xmin)/(nx-1)
fx = xmin
sx = Sampling(nx,dx,fx)
yg = BilateralFilter.sampleRangeFunction(gauss,sigma,sx)
yt = BilateralFilter.sampleRangeFunction(tukey,sigma,sx)
#yh = BilateralFilter.sampleRangeFunction(huber,sigma,sx)
#x = rampfloat(fx,dx,nx)
#yg = mul(x,yg)
#yh = mul(x,yh)
#yt = mul(x,yt)
sp = SimplePlot()
sp.setBackground(backgroundColor)
sp.setFontSizeForSlide(1.0,0.9)
sp.setSize(700,500)
sp.setHLabel(" ")
sp.setVLabel(" ")
solid = PointsView.Line.SOLID
dash = PointsView.Line.DASH
dot = PointsView.Line.DOT
#pv = sp.addPoints(sx,yh); pv.setLineStyle(dot); pv.setLineWidth(4)
pv = sp.addPoints(sx,yg); pv.setLineStyle(dash); pv.setLineWidth(4)
pv = sp.addPoints(sx,yt); pv.setLineStyle(solid); pv.setLineWidth(4)
if pngDir:
sp.paintToPng(720,3.3,pngDir+"blftg.png")
def bilateralFilterQC(sigmaS, sigmaX, t, x):
y = like(x)
bf = BilateralFilter(sigmaS, sigmaX)
if t:
qc = bf.applyQC(t, x, y)
else:
qc = bf.applyQC(x, y)
return qc.sx, qc.sn, qc.sd, qc.tn, qc.td
def bilateralFilterQC(sigmaS,sigmaX,t,x):
y = like(x)
bf = BilateralFilter(sigmaS,sigmaX)
if t:
qc = bf.applyQC(t,x,y)
else:
qc = bf.applyQC(x,y)
return qc.sx,qc.sn,qc.sd,qc.tn,qc.td
def bilateralFilter(sigmaS,sigmaX,t,x):
y = like(x)
bf = BilateralFilter(sigmaS,sigmaX)
if t:
bf.apply(t,x,y)
else:
bf.apply(x,y)
return y
def goFilterImpulses():
xa,s1,s2,clip = getImage()
xb = makeImpulses(12,len(xa[0]),len(xa))
t = diffusionTensors(2.0,xa)
plot(xa,s1,s2,1.3)
#plot(xb,s1,s2,0.9)
for sigmaR in [0.1,1.0,100.0]:
#for sigmaR in [100.0]:
y = like(xa)
bf = BilateralFilter(30.0,sigmaR)
bf.setType(BilateralFilter.Type.TUKEY)
bf.applyAB(t,xa,xb,y)
y = smoothS(y)
#plot(y,s1,s2,0.5*max(y))
plot(y,s1,s2,0.1)
def goRandomBlocks():
n1 = 801
x = makeBlocks(n1)
plotB(x,x,-12,12,"rbx")
return
y = add(x,makeNoiseForBlocks(3141,8.0,3.0,n1))
plotB(x,y,-12,12,"rbxy")
sigmaS = 20.0
yqqd = qqd(y)
z = zerofloat(n1)
for scale in [0.01,0.5,1.0,1.5,10,100.0]:
sigmaX = scale*yqqd
print "sigmaS =",sigmaS," sigmaX =",sigmaX
bf = BilateralFilter(sigmaS,sigmaX)
bf.setType(BilateralFilter.Type.TUKEY)
bf.apply(y,z)
png = "rbxz"+str(int(scale*10+0.5))
if len(png)==1: png = "0"+png
plotB(x,z,-12,12,png=png)
def goApproximation():
factor = 0.67448 # 3rd quartile of standard normal distribution (sigma = 1)
sigma = 1.401114
pmin, pmax = -9.0, 9.0
np, dp, fp = 301, 0.06, pmin
nk, dk, fk = 13, 1.5, pmin
sp = Sampling(np, dp, fp)
sk = Sampling(nk, dk, fk)
type = BilateralFilter.Type.GAUSS
rf = BilateralFilter.sampleRangeFunction(type, sigma, sp)
rg = BilateralFilter.sampleRangeFunction(type, sigma / sqrt(2.0), sp)
method = CubicInterpolator.Method.MONOTONIC
rfi = CubicInterpolator(method, np, rampfloat(fp, dp, np), rf)
rgi = CubicInterpolator(method, np, rampfloat(fp, dp, np), rg)
rr = zerofloat(np, np)
ra = zerofloat(np, np)
r0 = zerofloat(np, np)
rh = zerofloat(np, np)
for i in range(np):
pi = sp.getValue(i)
for j in range(np):
pj = sp.getValue(j)
rr[i][j] = rfi.interpolate(pi - pj)
for k in range(nk):
pk = sk.getValue(k)
rjk = rgi.interpolate(pj - pk)
rik = rgi.interpolate(pi - pk)
#hik = hat(dk,pi-pk)
#ra[i][j] += hik*rjk
ra[i][j] += rik * rjk
if k == nk / 2:
r0[i][j] = rjk
rh[i][j] = rik * rjk
#rh[i][j] = hik*rjk
rr = div(rr, max(rr))
r0 = div(r0, max(r0))
rh = div(rh, max(rh))
ra = div(ra, max(ra))
plotR2(sp, rr, "blfrr")
plotR2(sp, r0, "blfr0")
plotR2(sp, rh, "blfrh")
plotR2(sp, ra, "blfra")
def goApproximation():
factor = 0.67448 # 3rd quartile of standard normal distribution (sigma = 1)
sigma = 1.401114
pmin,pmax = -9.0,9.0
np,dp,fp = 301,0.06,pmin
nk,dk,fk = 13,1.5,pmin
sp = Sampling(np,dp,fp)
sk = Sampling(nk,dk,fk)
type = BilateralFilter.Type.GAUSS
rf = BilateralFilter.sampleRangeFunction(type,sigma,sp)
rg = BilateralFilter.sampleRangeFunction(type,sigma/sqrt(2.0),sp)
method = CubicInterpolator.Method.MONOTONIC
rfi = CubicInterpolator(method,np,rampfloat(fp,dp,np),rf)
rgi = CubicInterpolator(method,np,rampfloat(fp,dp,np),rg)
rr = zerofloat(np,np)
ra = zerofloat(np,np)
r0 = zerofloat(np,np)
rh = zerofloat(np,np)
for i in range(np):
pi = sp.getValue(i)
for j in range(np):
pj = sp.getValue(j)
rr[i][j] = rfi.interpolate(pi-pj)
for k in range(nk):
pk = sk.getValue(k)
rjk = rgi.interpolate(pj-pk)
rik = rgi.interpolate(pi-pk)
#hik = hat(dk,pi-pk)
#ra[i][j] += hik*rjk
ra[i][j] += rik*rjk
if k==nk/2:
r0[i][j] = rjk
rh[i][j] = rik*rjk
#rh[i][j] = hik*rjk
rr = div(rr,max(rr))
r0 = div(r0,max(r0))
rh = div(rh,max(rh))
ra = div(ra,max(ra))
plotR2(sp,rr,"blfrr")
plotR2(sp,r0,"blfr0")
plotR2(sp,rh,"blfrh")
plotR2(sp,ra,"blfra")
def bilateralFilter(sigmaS, sigmaX, t, x):
y = like(x)
bf = BilateralFilter(sigmaS, sigmaX)
if t:
bf.apply(t, x, y)
else:
bf.apply(x, y)
return y
def goOnce(v1,v2): t11,t12,t22 = like(g11),like(g12),like(g22) v = getAnglesFromGradients(v1,v2) sigmaS = 24.0 # spatial filter sigma sigmaR = computeSigmaR(v1) blf = BilateralFilter(sigmaS,sigmaR) blf.setType(BilateralFilter.Type.TUKEY_ANGLE) blf.applyABC(g,v,g11,t11) blf.applyABC(g,v,g12,t12) blf.applyABC(g,v,g22,t22) u1,u2,_,_,_,_ = getEigenFromTensors(t11,t12,t22) return u1,u2
def goFilterRandom():
xa, s1, s2, clip = getImage()
plot(xa, s1, s2, clip)
n1, n2 = len(xa[0]), len(xa)
xb = makeRandom(n1, n2)
t = diffusionTensors(2.0, xa)
#plot(xa,s1,s2,1.3)
#plot(xb,s1,s2,0.5)
xqqd = qqd(x)
for sigmaR in [xqqd, 100 * xqqd]:
y = like(xa)
bf = BilateralFilter(30.0, sigmaR)
bf.setType(BilateralFilter.Type.TUKEY)
bf.applyAB(t, xa, xb, y)
plot(y, s1, s2, 0.1)
bf.apply(t, xa, y)
plot(y, s1, s2, clip)
def goFilterRandom():
xa,s1,s2,clip = getImage()
plot(xa,s1,s2,clip)
n1,n2 = len(xa[0]),len(xa)
xb = makeRandom(n1,n2)
t = diffusionTensors(2.0,xa)
#plot(xa,s1,s2,1.3)
#plot(xb,s1,s2,0.5)
xqqd = qqd(x)
for sigmaR in [xqqd,100*xqqd]:
y = like(xa)
bf = BilateralFilter(30.0,sigmaR)
bf.setType(BilateralFilter.Type.TUKEY)
bf.applyAB(t,xa,xb,y)
plot(y,s1,s2,0.1)
bf.apply(t,xa,y)
plot(y,s1,s2,clip)
def goFilterImpulses():
xa, s1, s2, clip = getImage()
xb = makeImpulses(12, len(xa[0]), len(xa))
t = diffusionTensors(2.0, xa)
plot(xa, s1, s2, 1.3)
#plot(xb,s1,s2,0.9)
for sigmaR in [0.1, 1.0, 100.0]:
#for sigmaR in [100.0]:
y = like(xa)
bf = BilateralFilter(30.0, sigmaR)
bf.setType(BilateralFilter.Type.TUKEY)
bf.applyAB(t, xa, xb, y)
y = smoothS(y)
#plot(y,s1,s2,0.5*max(y))
plot(y, s1, s2, 0.1)
def goRandomBlocks():
n1 = 801
x = makeBlocks(n1)
plotB(x, x, -12, 12, "rbx")
return
y = add(x, makeNoiseForBlocks(3141, 8.0, 3.0, n1))
plotB(x, y, -12, 12, "rbxy")
sigmaS = 20.0
yqqd = qqd(y)
z = zerofloat(n1)
for scale in [0.01, 0.5, 1.0, 1.5, 10, 100.0]:
sigmaX = scale * yqqd
print "sigmaS =", sigmaS, " sigmaX =", sigmaX
bf = BilateralFilter(sigmaS, sigmaX)
bf.setType(BilateralFilter.Type.TUKEY)
bf.apply(y, z)
png = "rbxz" + str(int(scale * 10 + 0.5))
if len(png) == 1: png = "0" + png
plotB(x, z, -12, 12, png=png)
