qurator-spk
/
eynollah
mirror of https://github.com/qurator-spk/eynollah.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

🔥 manually merge 58ecb57

Browse Source
pull/8/head
Konstantin Baierer 4 years ago
parent 145b2fcd98
commit 6ab7abdfcd
2 changed files with 832 additions and 842 deletions
Show Stats Download Patch File Download Diff File

126
sbb_newspapers_org_image/eynollah.py
Unescape Escape View File

@ -1239,8 +1239,14 @@ class eynollah:
# all_box_coord.append(crop_coor)
mask_textline=np.zeros((textline_mask_tot_ea.shape))
mask_textline=cv2.fillPoly(mask_textline,pts=[contours_per_process[mv]],color=(1,1,1))
denoised = None
all_text_region_raw = textline_mask_tot_ea[boxes_text[mv][1] : boxes_text[mv][1] + boxes_text[mv][3], boxes_text[mv][0] : boxes_text[mv][0] + boxes_text[mv][2]]
all_text_region_raw=(textline_mask_tot_ea*mask_textline[:,:])[boxes_text[mv][1]:boxes_text[mv][1]+boxes_text[mv][3] , boxes_text[mv][0]:boxes_text[mv][0]+boxes_text[mv][2] ]
all_text_region_raw = all_text_region_raw.astype(np.uint8)
img_int_p = all_text_region_raw[:, :] # self.all_text_region_raw[mv]
@ -2762,15 +2768,19 @@ class eynollah:
patches=True
binary=False
ratio_y=1.3
ratio_x=1
median_blur=False
img = resize_image(img_org, int(img_org.shape[0] * ratio_y), int(img_org.shape[1] * ratio_x))
img= self.resize_image(img_org, int(img_org.shape[0]*ratio_y), int(img_org.shape[1]*ratio_x))
if binary:
img = otsu_copy_binary(img) # otsu_copy(img)
img = self.otsu_copy_binary(img)#self.otsu_copy(img)
img = img.astype(np.uint16)
if median_blur:
@ -2780,15 +2790,20 @@ class eynollah:
img = img.astype(np.uint16)
prediction_regions_org_y=self.do_prediction(patches,img,model_region)
prediction_regions_org_y = resize_image(prediction_regions_org_y, img_height_h, img_width_h)
prediction_regions_org_y=self.resize_image(prediction_regions_org_y, img_height_h, img_width_h )
#plt.imshow(prediction_regions_org_y[:,:,0])
#plt.show()
#sys.exit()
prediction_regions_org_y=prediction_regions_org_y[:,:,0]
mask_zeros_y=(prediction_regions_org_y[:,:]==0)*1
if is_image_enhanced:
ratio_x=1.2
else:
@ -2797,10 +2812,10 @@ class eynollah:
ratio_y=1
median_blur=False
img = resize_image(img_org, int(img_org.shape[0] * ratio_y), int(img_org.shape[1] * ratio_x))
img= self.resize_image(img_org, int(img_org.shape[0]*ratio_y), int(img_org.shape[1]*ratio_x))
if binary:
img = otsu_copy_binary(img) # otsu_copy(img)
img = self.otsu_copy_binary(img)#self.otsu_copy(img)
img = img.astype(np.uint16)
if median_blur:
@ -2810,7 +2825,7 @@ class eynollah:
img = img.astype(np.uint16)
prediction_regions_org=self.do_prediction(patches,img,model_region)
prediction_regions_org = resize_image(prediction_regions_org, img_height_h, img_width_h)
prediction_regions_org=self.resize_image(prediction_regions_org, img_height_h, img_width_h )
##plt.imshow(prediction_regions_org[:,:,0])
##plt.show()
@ -2830,14 +2845,17 @@ class eynollah:
patches=True
binary=False
ratio_x=1
ratio_y=1
median_blur=False
img = resize_image(img_org, int(img_org.shape[0] * ratio_y), int(img_org.shape[1] * ratio_x))
img= self.resize_image(img_org, int(img_org.shape[0]*ratio_y), int(img_org.shape[1]*ratio_x))
if binary:
img = otsu_copy_binary(img) # otsu_copy(img)
img = self.otsu_copy_binary(img)#self.otsu_copy(img)
img = img.astype(np.uint16)
if median_blur:
@ -2845,9 +2863,11 @@ class eynollah:
if gaussian_filter:
img= cv2.GaussianBlur(img,(5,5),0)
img = img.astype(np.uint16)
prediction_regions_org2 = self.do_prediction(patches, img, model_region)
prediction_regions_org2 = resize_image(prediction_regions_org2, img_height_h, img_width_h)
marginal_patch=0.2
prediction_regions_org2=self.do_prediction(patches,img,model_region,marginal_patch)
prediction_regions_org2=self.resize_image(prediction_regions_org2, img_height_h, img_width_h )
#plt.imshow(prediction_regions_org2[:,:,0])
#plt.show()
@ -2865,15 +2885,17 @@ class eynollah:
text_sume_early=( (prediction_regions_org[:,:]==1)*1 ).sum()
prediction_regions_org_copy=np.copy(prediction_regions_org)
prediction_regions_org_copy[(prediction_regions_org_copy[:,:]==1) & (mask_zeros2[:,:]==1)]=0
text_sume_second=( (prediction_regions_org_copy[:,:]==1)*1 ).sum()
rate_two_models=text_sume_second/float(text_sume_early)*100
print(rate_two_models, "ratio_of_two_models")
print(rate_two_models,'ratio_of_two_models')
if is_image_enhanced and rate_two_models<95.50:#98.45:
pass
else:
@ -2882,6 +2904,7 @@ class eynollah:
##prediction_regions_org[mask_lines2[:,:]==1]=3
prediction_regions_org[(mask_lines2[:,:]==1) & (prediction_regions_org[:,:]==0)]=3
del mask_lines2
del mask_zeros2
del prediction_regions_org2
@ -2895,14 +2918,17 @@ class eynollah:
#patches=True
#binary=False
#ratio_x=1
#ratio_y=1
#median_blur=False
# img= resize_image(img_org, int(img_org.shape[0]*ratio_y), int(img_org.shape[1]*ratio_x))
#img= self.resize_image(img_org, int(img_org.shape[0]*ratio_y), int(img_org.shape[1]*ratio_x))
#if binary:
# img = otsu_copy_binary(img)#otsu_copy(img)
#img = self.otsu_copy_binary(img)#self.otsu_copy(img)
#img = img.astype(np.uint16)
#if median_blur:
@ -2912,7 +2938,7 @@ class eynollah:
#img = img.astype(np.uint16)
#prediction_regions_org2=self.do_prediction(patches,img,model_region)
# prediction_regions_org2=resize_image(prediction_regions_org2, img_height_h, img_width_h )
#prediction_regions_org2=self.resize_image(prediction_regions_org2, img_height_h, img_width_h )
##plt.imshow(prediction_regions_org2[:,:,0])
##plt.show()
@ -2930,6 +2956,7 @@ class eynollah:
#text_sume_early=( (prediction_regions_org[:,:]==1)*1 ).sum()
#prediction_regions_org[(prediction_regions_org[:,:]==1) & (mask_zeros2[:,:]==1)]=0
###prediction_regions_org[mask_lines2[:,:]==1]=3
@ -2956,13 +2983,16 @@ class eynollah:
mask_images_only=(prediction_regions_org[:,:]==2)*1
pixel_img=1
min_area_text=0.00001
polygons_of_only_texts = return_contours_of_interested_region(mask_texts_only, pixel_img, min_area_text)
polygons_of_only_texts=self.return_contours_of_interested_region(mask_texts_only,pixel_img,min_area_text)
polygons_of_only_images = return_contours_of_interested_region(mask_images_only, pixel_img)
polygons_of_only_images=self.return_contours_of_interested_region(mask_images_only,pixel_img)
polygons_of_only_lines=self.return_contours_of_interested_region(mask_lines_only,pixel_img,min_area_text)
polygons_of_only_lines = return_contours_of_interested_region(mask_lines_only, pixel_img, min_area_text)
text_regions_p_true=np.zeros(prediction_regions_org.shape)
#text_regions_p_true[:,:]=text_regions_p_1[:,:]
@ -2976,6 +3006,7 @@ class eynollah:
##print(np.unique(text_regions_p_true))
#text_regions_p_true_3d=np.repeat(text_regions_p_1[:, :, np.newaxis], 3, axis=2)
#text_regions_p_true_3d=text_regions_p_true_3d.astype(np.uint8)
@ -2993,6 +3024,7 @@ class eynollah:
return text_regions_p_true
def write_images_into_directory(self, img_contoures, dir_of_cropped_imgs, image_page):
index = 0
for cont_ind in img_contoures:
@ -3013,6 +3045,7 @@ class eynollah:
mask_marginals=np.zeros((text_with_lines.shape[0],text_with_lines.shape[1]))
mask_marginals=mask_marginals.astype(np.uint8)
text_with_lines=text_with_lines.astype(np.uint8)
##text_with_lines=cv2.erode(text_with_lines,self.kernel,iterations=3)
@ -3021,13 +3054,14 @@ class eynollah:
if text_with_lines.shape[0]<=1500:
pass
elif text_with_lines.shape[0]>1500 and text_with_lines.shape[0]<=1800:
text_with_lines = resize_image(text_with_lines, int(text_with_lines.shape[0] * 1.5), text_with_lines.shape[1])
text_with_lines=self.resize_image(text_with_lines,int(text_with_lines.shape[0]*1.5),text_with_lines.shape[1])
text_with_lines=cv2.erode(text_with_lines,self.kernel,iterations=5)
text_with_lines = resize_image(text_with_lines, text_with_lines_eroded.shape[0], text_with_lines_eroded.shape[1])
text_with_lines=self.resize_image(text_with_lines,text_with_lines_eroded.shape[0],text_with_lines_eroded.shape[1])
else:
text_with_lines = resize_image(text_with_lines, int(text_with_lines.shape[0] * 1.8), text_with_lines.shape[1])
text_with_lines=self.resize_image(text_with_lines,int(text_with_lines.shape[0]*1.8),text_with_lines.shape[1])
text_with_lines=cv2.erode(text_with_lines,self.kernel,iterations=7)
text_with_lines = resize_image(text_with_lines, text_with_lines_eroded.shape[0], text_with_lines_eroded.shape[1])
text_with_lines=self.resize_image(text_with_lines,text_with_lines_eroded.shape[0],text_with_lines_eroded.shape[1])
text_with_lines_y=text_with_lines.sum(axis=0)
text_with_lines_y_eroded=text_with_lines_eroded.sum(axis=0)
@ -3043,15 +3077,21 @@ class eynollah:
else:
min_textline_thickness=40
if thickness_along_y_percent>=14:
text_with_lines_y_rev=-1*text_with_lines_y[:]
#print(text_with_lines_y)
#print(text_with_lines_y_rev)
#plt.plot(text_with_lines_y)
#plt.show()
text_with_lines_y_rev=text_with_lines_y_rev-np.min(text_with_lines_y_rev)
#plt.plot(text_with_lines_y_rev)
@ -3065,24 +3105,32 @@ class eynollah:
#plt.show()
region_sum_0_updown=region_sum_0[len(region_sum_0)::-1]
first_nonzero = next((i for i, x in enumerate(region_sum_0) if x), None)
last_nonzero = next((i for i, x in enumerate(region_sum_0_updown) if x), None)
first_nonzero=(next((i for i, x in enumerate(region_sum_0) if x), None))
last_nonzero=(next((i for i, x in enumerate(region_sum_0_updown) if x), None))
last_nonzero=len(region_sum_0)-last_nonzero
##img_sum_0_smooth_rev=-region_sum_0
mid_point = (last_nonzero + first_nonzero) / 2.0
mid_point=(last_nonzero+first_nonzero)/2.
one_third_right=(last_nonzero-mid_point)/3.0
one_third_left=(mid_point-first_nonzero)/3.0
#img_sum_0_smooth_rev=img_sum_0_smooth_rev-np.min(img_sum_0_smooth_rev)
peaks, _ = find_peaks(text_with_lines_y_rev, height=0)
peaks=np.array(peaks)
#print(region_sum_0[peaks])
##plt.plot(region_sum_0)
##plt.plot(peaks,region_sum_0[peaks],'*')
@ -3092,6 +3140,7 @@ class eynollah:
#print(first_nonzero,last_nonzero,peaks)
#print(region_sum_0[peaks]<10)
####peaks=peaks[region_sum_0[peaks]<25 ]
@ -3107,16 +3156,21 @@ class eynollah:
peaks_right=peaks[peaks>(mid_point+one_third_right)]
peaks_left=peaks[peaks<(mid_point-one_third_left)]
try:
point_right=np.min(peaks_right)
except:
point_right=last_nonzero
try:
point_left=np.max(peaks_left)
except:
point_left=first_nonzero
#print(point_left,point_right)
#print(text_regions.shape)
if point_right>=mask_marginals.shape[1]:
@ -3128,7 +3182,7 @@ class eynollah:
mask_marginals[:,:]=1
#print(mask_marginals.shape,point_left,point_right,'nadosh')
mask_marginals_rotated = rotate_image(mask_marginals, -slope_deskew)
mask_marginals_rotated=self.rotate_image(mask_marginals,-slope_deskew)
#print(mask_marginals_rotated.shape,'nadosh')
mask_marginals_rotated_sum=mask_marginals_rotated.sum(axis=0)
@ -3146,6 +3200,7 @@ class eynollah:
if max_point_of_right_marginal>=text_regions.shape[1]:
max_point_of_right_marginal=text_regions.shape[1]-1
#print(np.min(index_x_interest) ,np.max(index_x_interest),'minmaxnew')
#print(mask_marginals_rotated.shape,text_regions.shape,'mask_marginals_rotated')
#plt.imshow(mask_marginals)
@ -3156,11 +3211,14 @@ class eynollah:
text_regions[(mask_marginals_rotated[:,:]!=1) & (text_regions[:,:]==1)]=4
#plt.imshow(text_regions)
#plt.show()
pixel_img=4
min_area_text=0.00001
polygons_of_marginals = return_contours_of_interested_region(text_regions, pixel_img, min_area_text)
polygons_of_marginals=self.return_contours_of_interested_region(text_regions,pixel_img,min_area_text)
cx_text_only, cy_text_only, x_min_text_only, x_max_text_only, y_min_text_only, y_max_text_only, y_cor_x_min_main = find_new_features_of_contoures(polygons_of_marginals)
cx_text_only,cy_text_only ,x_min_text_only,x_max_text_only, y_min_text_only ,y_max_text_only,y_cor_x_min_main=self.find_new_features_of_contoures(polygons_of_marginals)
text_regions[(text_regions[:,:]==4)]=1
@ -3173,8 +3231,8 @@ class eynollah:
x_width_mar=abs(x_min_text_only[i]-x_max_text_only[i])
y_height_mar=abs(y_min_text_only[i]-y_max_text_only[i])
# print(x_width_mar,y_height_mar,'y_height_mar')
if x_width_mar > 16 and y_height_mar / x_width_mar < 10:
#print(x_width_mar,y_height_mar,y_height_mar/x_width_mar,'y_height_mar')
if x_width_mar>16 and y_height_mar/x_width_mar<18:
marginlas_should_be_main_text.append(polygons_of_marginals[i])
if x_min_text_only[i]<(mid_point-one_third_left):
x_min_marginals_left_new=x_min_text_only[i]
@ -3194,6 +3252,9 @@ class eynollah:
if len(x_min_marginals_right)==0:
x_min_marginals_right=[text_regions.shape[1]-1]
#print(x_min_marginals_left[0],x_min_marginals_right[0],'margo')
#print(marginlas_should_be_main_text,'marginlas_should_be_main_text')
@ -3214,6 +3275,7 @@ class eynollah:
#plt.plot(peaks,region_sum_0[peaks],'*')
#plt.show()
#plt.imshow(text_regions)
#plt.show()
@ -4119,7 +4181,7 @@ class eynollah:
num_col = None
peaks_neg_fin = []
print(num_col, "num_colnum_col")
#print(num_col, "num_colnum_col")
if num_col is None:
txt_con_org = []
order_text_new = []
@ -4144,7 +4206,7 @@ class eynollah:
K.clear_session()
gc.collect()
print(np.unique(textline_mask_tot_ea[:, :]), "textline")
#print(np.unique(textline_mask_tot_ea[:, :]), "textline")
if self.dir_of_all is not None:

476
sbb_newspapers_org_image/utils.py
Unescape Escape View File

@ -1493,99 +1493,106 @@ def filter_small_drop_capitals_from_no_patch_layout(layout_no_patch, layout1):
def find_num_col_deskew(regions_without_seperators, sigma_, multiplier=3.8):
regions_without_seperators_0=regions_without_seperators[:,:].sum(axis=1)
meda_n_updown = regions_without_seperators_0[len(regions_without_seperators_0) :: -1]
##meda_n_updown=regions_without_seperators_0[len(regions_without_seperators_0)::-1]
first_nonzero = next((i for i, x in enumerate(regions_without_seperators_0) if x), 0)
last_nonzero = next((i for i, x in enumerate(meda_n_updown) if x), 0)
##first_nonzero=(next((i for i, x in enumerate(regions_without_seperators_0) if x), 0))
##last_nonzero=(next((i for i, x in enumerate(meda_n_updown) if x), 0))
##last_nonzero=len(regions_without_seperators_0)-last_nonzero
last_nonzero = len(regions_without_seperators_0) - last_nonzero
y=regions_without_seperators_0#[first_nonzero:last_nonzero]
y_help = np.zeros(len(y) + 20)
##y_help=np.zeros(len(y)+20)
y_help[10 : len(y) + 10] = y
##y_help[10:len(y)+10]=y
x = np.array(range(len(y)))
##x=np.array( range(len(y)) )
zneg_rev = -y_help + np.max(y_help)
zneg = np.zeros(len(zneg_rev) + 20)
zneg[10 : len(zneg_rev) + 10] = zneg_rev
##zneg_rev=-y_help+np.max(y_help)
##zneg=np.zeros(len(zneg_rev)+20)
##zneg[10:len(zneg_rev)+10]=zneg_rev
z=gaussian_filter1d(y, sigma_)
zneg = gaussian_filter1d(zneg, sigma_)
###zneg= gaussian_filter1d(zneg, sigma_)
peaks_neg, _ = find_peaks(zneg, height=0)
peaks, _ = find_peaks(z, height=0)
peaks_neg = peaks_neg - 10 - 10
###peaks_neg, _ = find_peaks(zneg, height=0)
###peaks, _ = find_peaks(z, height=0)
# print(np.std(z),'np.std(z)np.std(z)np.std(z)')
###peaks_neg=peaks_neg-10-10
##plt.plot(z)
##plt.show()
####print(np.std(z),'np.std(z)np.std(z)np.std(z)')
##plt.imshow(regions_without_seperators)
##plt.show()
"""
last_nonzero=last_nonzero-0#100
first_nonzero=first_nonzero+0#+100
#####plt.plot(z)
#####plt.show()
peaks_neg=peaks_neg[(peaks_neg>first_nonzero) & (peaks_neg<last_nonzero)]
#####plt.imshow(regions_without_seperators)
#####plt.show()
###"""
###last_nonzero=last_nonzero-0#100
###first_nonzero=first_nonzero+0#+100
peaks=peaks[(peaks>.06*regions_without_seperators.shape[1]) & (peaks<0.94*regions_without_seperators.shape[1])]
"""
interest_pos = z[peaks]
###peaks_neg=peaks_neg[(peaks_neg>first_nonzero) & (peaks_neg<last_nonzero)]
interest_pos = interest_pos[interest_pos > 10]
###peaks=peaks[(peaks>.06*regions_without_seperators.shape[1]) & (peaks<0.94*regions_without_seperators.shape[1])]
###"""
###interest_pos=z[peaks]
interest_neg = z[peaks_neg]
###interest_pos=interest_pos[interest_pos>10]
min_peaks_pos = np.mean(interest_pos)
min_peaks_neg = 0 # np.min(interest_neg)
###interest_neg=z[peaks_neg]
dis_talaei = (min_peaks_pos - min_peaks_neg) / multiplier
# print(interest_pos)
grenze = min_peaks_pos - dis_talaei # np.mean(y[peaks_neg[0]:peaks_neg[len(peaks_neg)-1]])-np.std(y[peaks_neg[0]:peaks_neg[len(peaks_neg)-1]])/2.0
###min_peaks_pos=np.mean(interest_pos)
###min_peaks_neg=0#np.min(interest_neg)
interest_neg_fin = interest_neg[(interest_neg < grenze)]
peaks_neg_fin = peaks_neg[(interest_neg < grenze)]
interest_neg_fin = interest_neg[(interest_neg < grenze)]
###dis_talaei=(min_peaks_pos-min_peaks_neg)/multiplier
####print(interest_pos)
###grenze=min_peaks_pos-dis_talaei#np.mean(y[peaks_neg[0]:peaks_neg[len(peaks_neg)-1]])-np.std(y[peaks_neg[0]:peaks_neg[len(peaks_neg)-1]])/2.0
"""
if interest_neg[0]<0.1:
interest_neg=interest_neg[1:]
if interest_neg[len(interest_neg)-1]<0.1:
interest_neg=interest_neg[:len(interest_neg)-1]
###interest_neg_fin=interest_neg[(interest_neg<grenze)]
###peaks_neg_fin=peaks_neg[(interest_neg<grenze)]
###interest_neg_fin=interest_neg[(interest_neg<grenze)]
###"""
###if interest_neg[0]<0.1:
###interest_neg=interest_neg[1:]
###if interest_neg[len(interest_neg)-1]<0.1:
###interest_neg=interest_neg[:len(interest_neg)-1]
min_peaks_pos=np.min(interest_pos)
min_peaks_neg=0#np.min(interest_neg)
###min_peaks_pos=np.min(interest_pos)
###min_peaks_neg=0#np.min(interest_neg)
dis_talaei=(min_peaks_pos-min_peaks_neg)/multiplier
grenze=min_peaks_pos-dis_talaei#np.mean(y[peaks_neg[0]:peaks_neg[len(peaks_neg)-1]])-np.std(y[peaks_neg[0]:peaks_neg[len(peaks_neg)-1]])/2.0
"""
# interest_neg_fin=interest_neg#[(interest_neg<grenze)]
# peaks_neg_fin=peaks_neg#[(interest_neg<grenze)]
# interest_neg_fin=interest_neg#[(interest_neg<grenze)]
num_col = (len(interest_neg_fin)) + 1
###dis_talaei=(min_peaks_pos-min_peaks_neg)/multiplier
###grenze=min_peaks_pos-dis_talaei#np.mean(y[peaks_neg[0]:peaks_neg[len(peaks_neg)-1]])-np.std(y[peaks_neg[0]:peaks_neg[len(peaks_neg)-1]])/2.0
###"""
####interest_neg_fin=interest_neg#[(interest_neg<grenze)]
####peaks_neg_fin=peaks_neg#[(interest_neg<grenze)]
####interest_neg_fin=interest_neg#[(interest_neg<grenze)]
p_l = 0
p_u = len(y) - 1
p_m = int(len(y) / 2.0)
p_g_l = int(len(y) / 3.0)
p_g_u = len(y) - int(len(y) / 3.0)
###num_col=(len(interest_neg_fin))+1
diff_peaks = np.abs(np.diff(peaks_neg_fin))
diff_peaks_annormal = diff_peaks[diff_peaks < 30]
###p_l=0
###p_u=len(y)-1
###p_m=int(len(y)/2.)
###p_g_l=int(len(y)/3.)
###p_g_u=len(y)-int(len(y)/3.)
###diff_peaks=np.abs( np.diff(peaks_neg_fin) )
###diff_peaks_annormal=diff_peaks[diff_peaks<30]
#print(len(interest_neg_fin),np.mean(interest_neg_fin))
return interest_neg_fin, np.std(z)
return np.std(z)#interest_neg_fin,np.std(z)
def return_hor_spliter_by_index_for_without_verticals(peaks_neg_fin_t, x_min_hor_some, x_max_hor_some):
# print(peaks_neg_fin_t,x_min_hor_some,x_max_hor_some)
@ -2343,20 +2350,22 @@ def return_regions_without_seperators(regions_pre):
def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=None, f_name=None):
if main_page and dir_of_all is not None:
plt.figure(figsize=(70,40))
plt.rcParams["font.size"] = "50"
plt.rcParams['font.size']='50'
plt.subplot(1,2,1)
plt.imshow(img_patch_org)
plt.subplot(1,2,2)
plt.plot(gaussian_filter1d(img_patch_org.sum(axis=1), 3),np.array(range(len(gaussian_filter1d(img_patch_org.sum(axis=1), 3)))),linewidth=8)
plt.xlabel("Density of textline prediction in direction of X axis", fontsize=60)
plt.ylabel("Height", fontsize=60)
plt.xlabel('Density of textline prediction in direction of X axis',fontsize=60)
plt.ylabel('Height',fontsize=60)
plt.yticks([0,len(gaussian_filter1d(img_patch_org.sum(axis=1), 3))])
plt.gca().invert_yaxis()
plt.savefig(os.path.join(dir_of_all, f_name + "_density_of_textline.png"))
plt.savefig(os.path.join(dir_of_all, f_name+'_density_of_textline.png'))
#print(np.max(img_patch_org.sum(axis=0)) ,np.max(img_patch_org.sum(axis=1)),'axislar')
#img_patch_org=resize_image(img_patch_org,int(img_patch_org.shape[0]*2.5),int(img_patch_org.shape[1]/2.5))
@ -2366,60 +2375,76 @@ def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=Non
img_int=np.zeros((img_patch_org.shape[0],img_patch_org.shape[1]))
img_int[:,:]=img_patch_org[:,:]#img_patch_org[:,:,0]
img_resized = np.zeros((int(img_int.shape[0] * (1.8)), int(img_int.shape[1] * (2.6))))
img_resized[int(img_int.shape[0] * (0.4)) : int(img_int.shape[0] * (0.4)) + img_int.shape[0], int(img_int.shape[1] * (0.8)) : int(img_int.shape[1] * (0.8)) + img_int.shape[1]] = img_int[:, :]
max_shape=np.max(img_int.shape)
img_resized=np.zeros((int( max_shape*(1.1) ) , int( max_shape*(1.1) ) ))
onset_x=int((img_resized.shape[1]-img_int.shape[1])/2.)
onset_y=int((img_resized.shape[0]-img_int.shape[0])/2.)
#img_resized=np.zeros((int( img_int.shape[0]*(1.8) ) , int( img_int.shape[1]*(2.6) ) ))
#img_resized[ int( img_int.shape[0]*(.4)):int( img_int.shape[0]*(.4))+img_int.shape[0] , int( img_int.shape[1]*(.8)):int( img_int.shape[1]*(.8))+img_int.shape[1] ]=img_int[:,:]
img_resized[ onset_y:onset_y+img_int.shape[0] , onset_x:onset_x+img_int.shape[1] ]=img_int[:,:]
#print(img_resized.shape,'img_resizedshape')
#plt.imshow(img_resized)
#plt.show()
if main_page and img_patch_org.shape[1]>img_patch_org.shape[0]:
#plt.imshow(img_resized)
#plt.show()
angels = np.array(
[
-45,
0,
45,
90,
]
) # np.linspace(-12,12,100)#np.array([0 , 45 , 90 , -45])
res = []
num_of_peaks = []
index_cor = []
angels=np.array([-45, 0 , 45 , 90 , ])#np.linspace(-12,12,100)#np.array([0 , 45 , 90 , -45])
#res=[]
#num_of_peaks=[]
#index_cor=[]
var_res=[]
indexer = 0
#indexer=0
for rot in angels:
img_rot = rotate_image(img_resized, rot)
img_rot=self.rotate_image(img_resized,rot)
#plt.imshow(img_rot)
#plt.show()
img_rot[img_rot!=0]=1
# res_me=np.mean(find_num_col_deskew(img_rot,sigma_des,2.0 ))
#res_me=np.mean(self.find_num_col_deskew(img_rot,sigma_des,2.0 ))
# neg_peaks,var_spectrum=find_num_col_deskew(img_rot,sigma_des,20.3 )
#neg_peaks,var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
#print(var_spectrum,'var_spectrum')
try:
neg_peaks, var_spectrum = find_num_col_deskew(img_rot, sigma_des, 20.3)
# print(rot,var_spectrum,'var_spectrum')
res_me = np.mean(neg_peaks)
if res_me == 0:
res_me = VERY_LARGE_NUMBER
else:
pass
var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
##print(rot,var_spectrum,'var_spectrum')
#res_me=np.mean(neg_peaks)
#if res_me==0:
#res_me=1000000000000000000000
#else:
#pass
res_num = len(neg_peaks)
#res_num=len(neg_peaks)
except:
res_me = VERY_LARGE_NUMBER
res_num = 0
#res_me=1000000000000000000000
#res_num=0
var_spectrum=0
if isNaN(res_me):
pass
else:
res.append(res_me)
#if self.isNaN(res_me):
#pass
#else:
#res.append( res_me )
#var_res.append(var_spectrum)
#num_of_peaks.append( res_num )
#index_cor.append(indexer)
#indexer=indexer+1
var_res.append(var_spectrum)
num_of_peaks.append(res_num)
index_cor.append(indexer)
indexer = indexer + 1
#index_cor.append(indexer)
#indexer=indexer+1
try:
var_res=np.array(var_res)
@ -2428,42 +2453,31 @@ def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=Non
except:
ang_int=0
angels=np.linspace(ang_int-22.5,ang_int+22.5,100)
res = []
num_of_peaks = []
index_cor = []
#res=[]
#num_of_peaks=[]
#index_cor=[]
var_res=[]
indexer = 0
for rot in angels:
img_rot = rotate_image(img_resized, rot)
img_rot=self.rotate_image(img_resized,rot)
##plt.imshow(img_rot)
##plt.show()
img_rot[img_rot!=0]=1
# res_me=np.mean(find_num_col_deskew(img_rot,sigma_des,2.0 ))
#res_me=np.mean(self.find_num_col_deskew(img_rot,sigma_des,2.0 ))
try:
neg_peaks, var_spectrum = find_num_col_deskew(img_rot, sigma_des, 20.3)
# print(indexer,'indexer')
res_me = np.mean(neg_peaks)
if res_me == 0:
res_me = VERY_LARGE_NUMBER
else:
pass
var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
res_num = len(neg_peaks)
except:
res_me = VERY_LARGE_NUMBER
res_num = 0
var_spectrum=0
if isNaN(res_me):
pass
else:
res.append(res_me)
var_res.append(var_spectrum)
num_of_peaks.append(res_num)
index_cor.append(indexer)
indexer = indexer + 1
try:
var_res=np.array(var_res)
@ -2478,55 +2492,40 @@ def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=Non
#plt.show()
angels=np.linspace(-12,12,100)#np.array([0 , 45 , 90 , -45])
res = []
num_of_peaks = []
index_cor = []
var_res=[]
indexer = 0
for rot in angels:
img_rot = rotate_image(img_resized, rot)
img_rot=self.rotate_image(img_resized,rot)
#plt.imshow(img_rot)
#plt.show()
img_rot[img_rot!=0]=1
# res_me=np.mean(find_num_col_deskew(img_rot,sigma_des,2.0 ))
#res_me=np.mean(self.find_num_col_deskew(img_rot,sigma_des,2.0 ))
# neg_peaks,var_spectrum=find_num_col_deskew(img_rot,sigma_des,20.3 )
#neg_peaks,var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
#print(var_spectrum,'var_spectrum')
try:
neg_peaks, var_spectrum = find_num_col_deskew(img_rot, sigma_des, 20.3)
# print(rot,var_spectrum,'var_spectrum')
res_me = np.mean(neg_peaks)
if res_me == 0:
res_me = VERY_LARGE_NUMBER
else:
pass
var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
res_num = len(neg_peaks)
except:
res_me = VERY_LARGE_NUMBER
res_num = 0
var_spectrum=0
if isNaN(res_me):
pass
else:
res.append(res_me)
var_res.append(var_spectrum)
num_of_peaks.append(res_num)
index_cor.append(indexer)
indexer = indexer + 1
if dir_of_all is not None:
print("galdi?")
if self.dir_of_all is not None:
#print('galdi?')
plt.figure(figsize=(60,30))
plt.rcParams["font.size"] = "50"
plt.plot(angels, np.array(var_res), "-o", markersize=25, linewidth=4)
plt.xlabel("angle", fontsize=50)
plt.ylabel("variance of sum of rotated textline in direction of x axis", fontsize=50)
plt.rcParams['font.size']='50'
plt.plot(angels,np.array(var_res),'-o',markersize=25,linewidth=4)
plt.xlabel('angle',fontsize=50)
plt.ylabel('variance of sum of rotated textline in direction of x axis',fontsize=50)
plt.plot(angels[np.argmax(var_res)],var_res[np.argmax(np.array(var_res))] ,'*',markersize=50,label='Angle of deskewing=' +str("{:.2f}".format(angels[np.argmax(var_res)]))+r'$\degree$')
plt.legend(loc='best')
plt.savefig(os.path.join(self.dir_of_all,self.f_name+'_rotation_angle.png'))
plt.plot(angels[np.argmax(var_res)], var_res[np.argmax(np.array(var_res))], "*", markersize=50, label="Angle of deskewing=" + str("{:.2f}".format(angels[np.argmax(var_res)])) + r"$\degree$")
plt.legend(loc="best")
plt.savefig(os.path.join(dir_of_all, f_name + "_rotation_angle.png"))
try:
var_res=np.array(var_res)
@ -2535,45 +2534,27 @@ def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=Non
except:
ang_int=0
early_slope_edge=11
if abs(ang_int)>early_slope_edge and ang_int<0:
angels=np.linspace(-90,-12,100)
res = []
num_of_peaks = []
index_cor = []
var_res=[]
indexer = 0
for rot in angels:
img_rot = rotate_image(img_resized, rot)
img_rot=self.rotate_image(img_resized,rot)
##plt.imshow(img_rot)
##plt.show()
img_rot[img_rot!=0]=1
# res_me=np.mean(find_num_col_deskew(img_rot,sigma_des,2.0 ))
#res_me=np.mean(self.find_num_col_deskew(img_rot,sigma_des,2.0 ))
try:
neg_peaks, var_spectrum = find_num_col_deskew(img_rot, sigma_des, 20.3)
# print(indexer,'indexer')
res_me = np.mean(neg_peaks)
if res_me == 0:
res_me = VERY_LARGE_NUMBER
else:
pass
res_num = len(neg_peaks)
var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
except:
res_me = VERY_LARGE_NUMBER
res_num = 0
var_spectrum=0
if isNaN(res_me):
pass
else:
res.append(res_me)
var_res.append(var_spectrum)
num_of_peaks.append(res_num)
index_cor.append(indexer)
indexer = indexer + 1
try:
var_res=np.array(var_res)
@ -2586,40 +2567,23 @@ def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=Non
angels=np.linspace(90,12,100)
res = []
num_of_peaks = []
index_cor = []
var_res=[]
indexer = 0
for rot in angels:
img_rot = rotate_image(img_resized, rot)
img_rot=self.rotate_image(img_resized,rot)
##plt.imshow(img_rot)
##plt.show()
img_rot[img_rot!=0]=1
# res_me=np.mean(find_num_col_deskew(img_rot,sigma_des,2.0 ))
#res_me=np.mean(self.find_num_col_deskew(img_rot,sigma_des,2.0 ))
try:
neg_peaks, var_spectrum = find_num_col_deskew(img_rot, sigma_des, 20.3)
var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
#print(indexer,'indexer')
res_me = np.mean(neg_peaks)
if res_me == 0:
res_me = VERY_LARGE_NUMBER
else:
pass
res_num = len(neg_peaks)
except:
res_me = VERY_LARGE_NUMBER
res_num = 0
var_spectrum=0
if isNaN(res_me):
pass
else:
res.append(res_me)
var_res.append(var_spectrum)
num_of_peaks.append(res_num)
index_cor.append(indexer)
indexer = indexer + 1
try:
var_res=np.array(var_res)
@ -2629,45 +2593,29 @@ def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=Non
ang_int=0
else:
angels=np.linspace(-25,25,60)
res = []
num_of_peaks = []
index_cor = []
var_res=[]
indexer=0
for rot in angels:
img_rot = rotate_image(img_resized, rot)
img_rot=self.rotate_image(img_resized,rot)
#plt.imshow(img_rot)
#plt.show()
img_rot[img_rot!=0]=1
# res_me=np.mean(find_num_col_deskew(img_rot,sigma_des,2.0 ))
#res_me=np.mean(self.find_num_col_deskew(img_rot,sigma_des,2.0 ))
# neg_peaks,var_spectrum=find_num_col_deskew(img_rot,sigma_des,20.3 )
#neg_peaks,var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
#print(var_spectrum,'var_spectrum')
try:
neg_peaks, var_spectrum = find_num_col_deskew(img_rot, sigma_des, 20.3)
# print(rot,var_spectrum,'var_spectrum')
res_me = np.mean(neg_peaks)
if res_me == 0:
res_me = VERY_LARGE_NUMBER
else:
pass
res_num = len(neg_peaks)
var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
except:
res_me = VERY_LARGE_NUMBER
res_num = 0
var_spectrum=0
if isNaN(res_me):
pass
else:
res.append(res_me)
var_res.append(var_spectrum)
num_of_peaks.append(res_num)
index_cor.append(indexer)
indexer = indexer + 1
try:
var_res=np.array(var_res)
@ -2676,47 +2624,35 @@ def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=Non
except:
ang_int=0
# print(ang_int,'ang_int')
#plt.plot(var_res)
#plt.show()
##plt.plot(mom3_res)
##plt.show()
#print(ang_int,'ang_int111')
early_slope_edge=22
if abs(ang_int)>early_slope_edge and ang_int<0:
angels=np.linspace(-90,-25,60)
res = []
num_of_peaks = []
index_cor = []
var_res=[]
indexer = 0
for rot in angels:
img_rot = rotate_image(img_resized, rot)
img_rot=self.rotate_image(img_resized,rot)
##plt.imshow(img_rot)
##plt.show()
img_rot[img_rot!=0]=1
# res_me=np.mean(find_num_col_deskew(img_rot,sigma_des,2.0 ))
#res_me=np.mean(self.find_num_col_deskew(img_rot,sigma_des,2.0 ))
try:
neg_peaks, var_spectrum = find_num_col_deskew(img_rot, sigma_des, 20.3)
# print(indexer,'indexer')
res_me = np.mean(neg_peaks)
if res_me == 0:
res_me = VERY_LARGE_NUMBER
else:
pass
var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
res_num = len(neg_peaks)
except:
res_me = VERY_LARGE_NUMBER
res_num = 0
var_spectrum=0
if isNaN(res_me):
pass
else:
res.append(res_me)
var_res.append(var_spectrum)
num_of_peaks.append(res_num)
index_cor.append(indexer)
indexer = indexer + 1
try:
var_res=np.array(var_res)
@ -2729,40 +2665,24 @@ def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=Non
angels=np.linspace(90,25,60)
res = []
num_of_peaks = []
index_cor = []
var_res=[]
indexer=0
for rot in angels:
img_rot = rotate_image(img_resized, rot)
img_rot=self.rotate_image(img_resized,rot)
##plt.imshow(img_rot)
##plt.show()
img_rot[img_rot!=0]=1
# res_me=np.mean(find_num_col_deskew(img_rot,sigma_des,2.0 ))
#res_me=np.mean(self.find_num_col_deskew(img_rot,sigma_des,2.0 ))
try:
neg_peaks, var_spectrum = find_num_col_deskew(img_rot, sigma_des, 20.3)
var_spectrum=self.find_num_col_deskew(img_rot,sigma_des,20.3 )
#print(indexer,'indexer')
res_me = np.mean(neg_peaks)
if res_me == 0:
res_me = VERY_LARGE_NUMBER
else:
pass
res_num = len(neg_peaks)
except:
res_me = VERY_LARGE_NUMBER
res_num = 0
var_spectrum=0
if isNaN(res_me):
pass
else:
res.append(res_me)
var_res.append(var_spectrum)
num_of_peaks.append(res_num)
index_cor.append(indexer)
indexer = indexer + 1
try:
var_res=np.array(var_res)
@ -2773,6 +2693,7 @@ def return_deskew_slop(img_patch_org, sigma_des, main_page=False, dir_of_all=Non
return ang_int
def put_drop_out_from_only_drop_model(layout_no_patch, layout1):
drop_only = (layout_no_patch[:, :, 0] == 4) * 1
@ -2826,11 +2747,11 @@ def putt_bb_of_drop_capitals_of_model_in_patches_in_layout(layout_in_patch):
return layout_in_patch
def check_any_text_region_in_model_one_is_main_or_header(regions_model_1,regions_model_full,contours_only_text_parent,all_box_coord,all_found_texline_polygons,slopes,contours_only_text_parent_d_ordered):
text_only = (regions_model_1[:, :] == 1) * 1
contours_only_text, hir_on_text = return_contours_of_image(text_only)
#text_only=(regions_model_1[:,:]==1)*1
#contours_only_text,hir_on_text=self.return_contours_of_image(text_only)
"""
contours_only_text_parent=return_parent_contours( contours_only_text,hir_on_text)
contours_only_text_parent=self.return_parent_contours( contours_only_text,hir_on_text)
areas_cnt_text=np.array([cv2.contourArea(contours_only_text_parent[j]) for j in range(len(contours_only_text_parent))])
areas_cnt_text=areas_cnt_text/float(text_only.shape[0]*text_only.shape[1])
@ -2842,11 +2763,13 @@ def check_any_text_region_in_model_one_is_main_or_header(regions_model_1, region
contours_only_text_parent=[contours_only_text_parent[jz] for jz in range(len(contours_only_text_parent)) if areas_cnt_text[jz]>0.00001]
"""
cx_main, cy_main, x_min_main, x_max_main, y_min_main, y_max_main, y_corr_x_min_from_argmin = find_new_features_of_contoures(contours_only_text_parent)
cx_main,cy_main ,x_min_main , x_max_main, y_min_main ,y_max_main,y_corr_x_min_from_argmin=self.find_new_features_of_contoures(contours_only_text_parent)
length_con=x_max_main-x_min_main
height_con=y_max_main-y_min_main
all_found_texline_polygons_main=[]
all_found_texline_polygons_head=[]
@ -2867,11 +2790,14 @@ def check_any_text_region_in_model_one_is_main_or_header(regions_model_1, region
img=np.zeros((regions_model_1.shape[0],regions_model_1.shape[1],3))
img = cv2.fillPoly(img, pts=[con], color=(255, 255, 255))
all_pixels=((img[:,:,0]==255)*1).sum()
pixels_header=( ( (img[:,:,0]==255) & (regions_model_full[:,:,0]==2) )*1 ).sum()
pixels_main=all_pixels-pixels_header
if (pixels_header>=pixels_main) and ( (length_con[ii]/float(height_con[ii]) )>=1.3 ):
regions_model_1[:,:][(regions_model_1[:,:]==1) & (img[:,:,0]==255) ]=2
contours_only_text_parent_head.append(con)
@ -2891,6 +2817,8 @@ def check_any_text_region_in_model_one_is_main_or_header(regions_model_1, region
#print(all_pixels,pixels_main,pixels_header)
#plt.imshow(img[:,:,0])
#plt.show()
return regions_model_1,contours_only_text_parent_main,contours_only_text_parent_head,all_box_coord_main,all_box_coord_head,all_found_texline_polygons_main,all_found_texline_polygons_head,slopes_main,slopes_head,contours_only_text_parent_main_d,contours_only_text_parent_head_d

Write Preview
Loading…
Cancel
Save