def full_demo_orient_field():
"""
Estimates orientation field of a random print from FVC2002-1A, and
displays the result as a vector field overlayed on the original print
image a la Feng 2012.
"""
img = data.FVC2002_1A.read_random_image()[0]
orf = model.estimate_orientation(img, True)[:,:,0]
# This will plot a red line in the direction of the detected orientation
# for each block, overlayed on the original print, just like the figures
# in Chen 2011. This should probably be a utility function.
for ii in range(orf.shape[0]):
for jj in range(orf.shape[1]):
i_base = 16*ii + 8
j_base = 16*jj + 8
angle = orf[ii][jj]
if angle == np.nan:
continue
else:
offset = cmath.rect(8.0, angle)
plt.plot([j_base-offset.imag, j_base+offset.imag],
[i_base-offset.real, i_base+offset.real],
color='r')
plt.imshow(img, cmap = "gray", interpolation = "nearest")
plt.show()
def show_generated_img(_):
global canvas
global olpped_img
global olpped_pht
global show_step
global demo_dat
global orf
if show_step == "SHOW_OVERLAPPED":
print "SHOW OVERLAPPED GEN"
lhs_img_src = src_rgbt_imgs['CURLHS']
rhs_img_src = src_rgbt_imgs['CURRHS']
if (lhs_img_src == 'NONE') or (rhs_img_src == 'NONE'):
print "LEAVING EARLY FROM SHOW GEN IMG"
return None
canvas.itemconfig(lhs_img_id, state=TK.HIDDEN)
canvas.itemconfig(rhs_img_id, state=TK.HIDDEN)
gen_img_lhs = PIL.Image.new('L', (CANVAS_WIDTH, CANVAS_HEIGHT), 255)
# olpped_img_data["gen_img_lhs"] = gen_img_lhs
gen_img_rhs = PIL.Image.new('L', (CANVAS_WIDTH, CANVAS_HEIGHT), 255)
# olpped_img_data["gen_img_rhs"] = gen_img_rhs
lhs_img = src_gray_imgs['CURLHS']
# olpped_img_data["lhs_img"] = lhs_img
rhs_img = src_gray_imgs['CURRHS']
# olpped_img_data["rhs_img"] = rhs_img
gen_img_lhs.paste(lhs_img, (50, 0))
gen_img_rhs.paste(rhs_img, (CANVAS_WIDTH-IMG_WIDTH-50, 0))
eval_exp = "convert(min(a,b), 'L')"
olpped_img = PIL.ImageMath.eval(eval_exp, a=gen_img_lhs, b=gen_img_rhs)
demo_dat['olpped_img'] = olpped_img
# a=gen_img_lhs.convert('1', dither=0),\
# b=gen_img_rhs.convert('1', dither=0))
olpped_pht = PIL.ImageTk.PhotoImage(olpped_img)
canvas.itemconfig(olpped_img_id, image=olpped_pht, state=TK.NORMAL)
prnt_img_lhs_ary = np.array(gen_img_lhs, copy=True)
demo_dat['prnt_img_lhs_ary'] = prnt_img_lhs_ary
prnt_img_rhs_ary = np.array(gen_img_rhs, copy=True)
demo_dat['prnt_img_rhs_ary'] = prnt_img_rhs_ary
demo_dat['prnt_layers'] = \
np.dstack((prnt_img_lhs_ary, prnt_img_rhs_ary))
reg_lhs_prnt = utils.extract_region_chull(prnt_img_lhs_ary)
demo_dat['reg_lhs_prnt'] = reg_lhs_prnt
reg_rhs_prnt = utils.extract_region_chull(prnt_img_rhs_ary)
demo_dat['reg_rhs_prnt'] = reg_rhs_prnt
plhs = PIL.Image.fromarray(reg_lhs_prnt.astype(np.uint8) * 255, \
mode='L')
prhs = PIL.Image.fromarray(reg_rhs_prnt.astype(np.uint8) * 255, \
mode='L')
prnt_reg_img_rgb = PIL.Image.merge('RGB', [plhs,prhs,olpped_img])
demo_dat['prnt_reg_img_rgb'] = prnt_reg_img_rgb.copy()
prnt_reg_pht_rgb = PIL.ImageTk.PhotoImage(prnt_reg_img_rgb.copy())
# canvas.itemconfig(olpped_img_id, image=prnt_reg_pht_rgb)
show_step = "PRINT_REGIONS"
## ==== SHOW PRINT REGIONS
elif show_step == "PRINT_REGIONS":
print "SHOW PRINT_REGIONS"
prnt_reg_pht_rgb = PIL.ImageTk.PhotoImage(demo_dat['prnt_reg_img_rgb'])
demo_dat['prnt_reg_pht_rgb'] = prnt_reg_pht_rgb
canvas.itemconfig(olpped_img_id, image=prnt_reg_pht_rgb)
olpped_ary = np.array(demo_dat['olpped_img'], dtype=np.uint8, \
copy=True)
show_step = "OR_FIELD"
elif show_step == "OR_FIELD":
either_reg = demo_dat["reg_rhs_prnt"] | demo_dat["reg_lhs_prnt"]
tka_frame = TK.Frame(root, bd=-2, background="white", padx=0, pady=0)
tka_frame.grid(column=1, row=0, rowspan=5, sticky=TK.N+TK.S+TK.E+TK.W)
demo_dat['tka_frame'] = tka_frame
fig = Figure(figsize=(7, 5.5), facecolor="w", frameon=False, \
edgecolor='w')
demo_dat['fig'] = fig
ax = fig.add_axes((0,0,1,1), frameon=False, axis_bgcolor='w',\
axisbg='w')
demo_dat['ax'] = ax
tkagg = FigureCanvasTkAgg(fig, master=tka_frame)
demo_dat['tkagg'] = tkagg
tkagg.get_tk_widget().place(relx=0.5, rely=0.5, anchor=TK.CENTER)
tkagg.get_tk_widget().config(height=(CANVAS_HEIGHT+100), \
width=(CANVAS_WIDTH+150), bd=-2, \
bg='white', selectborderwidth=-2,
highlightthickness=0)
ax.axis('off')
fig.patch.set_visible(False)
canvas.delete(TK.ALL)
# canvas.place(relx=0.5, rely=0.5, anchor=TK.CENTER)
olpped_ary = np.array(demo_dat['olpped_img'], dtype=np.uint8, \
copy=True)
ax.imshow(olpped_ary, cmap="gray", interpolation="nearest")
either_reg = model._blockify_print_region(either_reg)
if 'orf' not in globals().keys():
orf = model.estimate_orientation(olpped_ary)
# demo_dat['orf'] = orf
for ii in range(orf.shape[0]):
for jj in range(orf.shape[1]):
angle = orf[ii, jj, 0]
if angle == np.nan or not (either_reg[ii,jj]):
continue
i_base = 16*ii + 8
j_base = 16*jj + 8
ioffst = 8*np.sin(angle+cmath.pi/2)
joffst = 8*np.cos(angle+cmath.pi/2)
ax.plot([j_base-joffst, j_base+joffst],
[i_base-ioffst, i_base+ioffst],
color='b', lw=1.5)
for ii in range(orf.shape[0]):
for jj in range(orf.shape[1]):
angle = orf[ii, jj, 1]
if angle == np.nan or not (either_reg[ii,jj]):
continue
i_base = 16*ii + 8
j_base = 16*jj + 8
ioffst = 8*np.sin(angle+cmath.pi/2)
joffst = 8*np.cos(angle+cmath.pi/2)
ax.plot([j_base-joffst, j_base+joffst],
[i_base-ioffst, i_base+ioffst],
color='r', lw=1.5)
ax.xaxis.set_visible(False)
ax.yaxis.set_visible(False)
tkagg.show()
show_step = "RELAX_LABEL_PROC"
if show_step == "RELAX_LABEL_PROC":
# orf = demo_dat['orf']
# oc = np.copy(orf)
pl_lhs = model._blockify_print_region(\
np.array(demo_dat["reg_lhs_prnt"], dtype=np.bool, copy=True))
pl_rhs = model._blockify_print_region( \
np.array(demo_dat["reg_rhs_prnt"], dtype=np.bool, copy=True))
p_lyrs = np.dstack((pl_lhs, pl_rhs))
tkagg = demo_dat['tkagg']
ax = demo_dat['ax']
print "orf.shape:" + str(orf.shape)
print "orf.dtype:" + str(orf.dtype)
print "p_lyrs.shape:" + str(p_lyrs.shape)
print "p_lyrs.dtype:" + str(p_lyrs.dtype)
_ = model.relax_label_two(orf, p_lyrs, tkagg, view_axis=ax)
