apply_deep_ocr — Apply a Deep OCR model on a set of images for inference.
apply_deep_ocr(Image : : DeepOcrHandle, Mode : DeepOcrResult)
apply_deep_ocr applies the Deep OCR model given by
DeepOcrHandle on the tuple of input images Image.
The operator returns DeepOcrResult, a tuple with a result
dictionary for every input image.
The operator apply_deep_ocr poses requirements on the input
Image:
Image type: byte.
Number of channels: 1 or 3.
Further, the operator apply_deep_ocr will preprocess the given
Image to match the model specifications.
This means, the byte images will be normalized and converted to type real.
Further, for Mode = 'auto' or 'detection'
the input image Image is padded to the model input dimensions and,
in case it has only one channel, converted into a three-channel image.
For Mode = 'recognition', three-channel images are
automatically converted to single-channel images.
The parameter Mode specifies a mode and with it, which component is
executed. Supported values:
Perform both parts, detection of the word and its recognition.
Perform only the detection part. Hence, the model will merely localize the word regions within the image.
Perform only the recognition part. Hence, the model requires that the image contains solely a tight crop of the word.
Note, the model must have been created with the desired component, see
create_deep_ocr.
The output dictionary DeepOcrResult can have entries according to the
applied Mode (marked by its abbreviation):
image (A, DET, REC):Preprocessed image.
score_maps (A, DET):Scores given as image with four channels:
Character score: Score for the character detection.
Link score: Score for the connection of detected character centers to a connected word.
Orientation 1: Sine component of the predicted word orientation.
Orientation 2: Cosine component of the predicted word orientation.
words (A, DET):Dictionary containing the following entries. Thereby, the entries are tuples with a value for every found word.
word (A): Recognized word.
char_candidates (A):
A dictionary with information for every character of every recognized
word. The dictionary contains for every word a key/value pair: The
index of the word as key and a tuple of dictionaries as value. Each of
these character dictionaries contains the following key/value
pairs:
'candidate': Tuple with the best
'recognition_num_char_candidates' candidates.
'confidence': Softmax based confidence values of the
best candidates. Note, these values are not calibrated and should
be used with care. They can vary significantly for different models.
word_image (A): Preprocessed image part containing the word.
row (A, DET): Localized word: Center point, row coordinate.
col (A, DET): Localized word: Center point, column
coordinate.
phi (A, DET): Localized word: Angle phi.
length1 (A, DET): Localized word: Half length of edge 1.
length2 (A, DET): Localized word: Half length of edge 2.
line_index (A, DET): Line index of localized word if
'detection_sort_by_line' set to 'true'.
The word localization is given by the parameters of an oriented rectangle,
see gen_rectangle2 for further information.
word_boxes_on_image (A, DET):
Dictionary with the word
localization on the coordinate system of the preprocessed images placed in
image. The entries are tuples with a value for every found word.
row (A, DET): Localized word: Center point, row coordinate.
col (A, DET): Localized word: Center point, column
coordinate.
phi (A, DET): Localized word: Angle phi.
length1 (A, DET): Localized word: Half length of edge 1.
length2 (A, DET): Localized word: Half length of edge 2.
The word localization is given by the parameters of an oriented rectangle,
see gen_rectangle2 for further information.
word_boxes_on_score_maps (A, DET):
Dictionary with the word
localization on the coordinate system of the score images placed in
score_maps. The entries are the same as for
word_boxes_on_image above.
word (REC):Recognized word.
char_candidates (REC):A tuple of dictionaries with information for every character in the recognized word.
Each of these character dictionaries contains the following key/value pairs:
'candidate': Tuple with the best
'recognition_num_char_candidates' candidates.
'confidence': Softmax based confidence values of the best
candidates. Note, these values are not calibrated and should be used
with care. They can vary significantly for different models.
The recognition component can be retrained with custom data in order to further enhance the performance. See OCR / Deep OCR for more information.
System requirements:
To run this operator on GPU (see get_deep_ocr_param), cuDNN and cuBLAS
are required.
For further details, please refer to the “Installation Guide”,
paragraph “Requirements for Deep Learning and Deep-Learning-Based Methods”.
Alternatively, this operator can also be run on CPU.
This operator returns a handle. Note that the state of an instance of this handle type may be changed by specific operators even though the handle is used as an input parameter by those operators.
This operator supports canceling timeouts and interrupts.
This operator supports breaking timeouts and interrupts.
Image (input_object) (multichannel-)image(-array) → object (byte)
Input image.
DeepOcrHandle (input_control) deep_ocr → (handle)
Handle of the Deep OCR model.
Mode (input_control) string → (string)
Inference mode.
Default: []
List of values: 'auto', 'detection', 'recognition'
DeepOcrResult (output_control) dict(-array) → (handle)
Tuple of result dictionaries.
If the parameters are valid, the operator apply_deep_ocr
returns the value 2 (
H_MSG_TRUE)
. If necessary, an exception is raised.
get_deep_ocr_param,
set_deep_ocr_param,
create_deep_ocr
OCR/OCV