Optical Armenian Handwritten Character Recognition - Demo - © Calfa.fr¶
Manuscript in Erkat‘agir
Out[126]:
Expected results¶(in red boxes)
In [127]:
page_gt.plot_image(with_char_boxes=True, figsize=(30,20), color='r', linewidth=2.5)
Automatic line detection¶
In [128]:
detected_lines = strategy_gradient(page_gt)
page_empty = page_gt.clone()
page_empty.lines = {i:line for i,line in enumerate(detected_lines)}
In [129]:
page_empty.plot_image(with_lines=True, figsize=(30,20), color='b', linewidth=2)
Load Convolutional Neural Network optimized for Line Transcription¶
In [130]:
model = models.line_transcriptors
In [131]:
len(page_empty.lines)
Out[131]:
In [132]:
line_id = 1
In [133]:
page_empty.lines[line_id].get_image()
Out[133]:
In [134]:
page_gt.lines = {i:page_gt.lines[line_id] for i,line_id in enumerate(sorted(page_gt.lines, key=lambda line_id: page_gt.lines[line_id].get_coord()['y']))}
Expected results¶
In [135]:
page_gt.lines[line_id].plot_image(with_char_boxes=True, figsize=(20,20), color='r', linewidth=3)
Single forward pass¶
In [136]:
line_transcripted = model(page_empty.lines[line_id])
Automatic Detection & Recognition results¶
In [137]:
line_transcripted
In [138]:
print('ground truth:', page_gt.lines[line_id].get_class_names())
print(' prediction:', line_transcripted.get_class_names())