Recognition settings

Aspose.OCR for C++ provides good recognition accuracy and performance by default. However, there will inevitably be cases where the default settings cannot provide reliable recognition results.

To configure Aspose.OCR recognition settings, pass an optional RecognitionSettings structure to the recognition function.

Setting	Type	Default value	Description
all_image	bool	false	Force recognition of the entire image. Not used for receipt recognition.
allowed_characters	characters_allowed_type	characters_allowed_type::ALL	The predefined whitelist of characters Aspose.OCR engine will look for.
alphabet	string	All symbols	A custom list of characters to be recognized, provided as a case-sensitive string. Characters that do not match the provided list are ignored.
auto_contrast	bool	false	Automatically increase the contrast of images before proceeding to recognition.
auto_denoising	bool	false	Automatically remove noise from images before proceeding to recognition.
correct_skew	bool	true	Automatically correct image tilt (deskew) before proceeding to recognition.
detect_areas_mode	detect_areas_mode_enum	detect_areas_mode_enum::DOCUMENT	Manually override the default document areas detection function. Not used for receipt recognition.
filters	custom_preprocessing_filters	none	Preprocessing filters to be applied to the image.
ignoredCharacters	string	none	A blacklist of characters that are ignored during recognition.
language_alphabet	language	language::none	Specify a language for recognition.
lines_filtration	bool	false	Set to true to recognize text in tables. Set to false to improve performance by ignoring table structures and treating tables as plain text.
preprocess_area	rect*	NULL	Specify the image region to be affected by preprocessing filters.
rectangles	rect*	NULL	Areas of the image from which to extract text.
rectangles_size	size_t	0	The number of recognition areas.
save_format	file_format	file_format::txt	The file format in which the recognition result is saved.
skew	double	0	Manually rotate the image by the specified degree. Does not work if recognition areas are specified.
threshold_value	int	0	Override the automatic binarization settings.
upscale_small_font	bool	false	Improve small font recognition and detection of dense lines.
defect_type	int	-1	The type of image defects that the OCR engine will look for.

Detected image defects

The following types of image defects can be detected:

Enum	Value	Behavior
ASPOSE_OCR_NONE	0	Do not detect image defects (default).
ASPOSE_OCR_DETECT_DARK_IMAGES	1	Detect low contrast between text and background. Such areas may not be recognized accurately, and their content may even disappear in the recognition results.
ASPOSE_OCR_DETECT_SALT_PEPPER_NOISE	2	Detect salt-and-pepper noise (impulse noise) that often occurs in digital photographs taken in low light conditions. It appears as random white and black pixels. This type of noise can cause certain characters to be misidentified or non-existent characters such as dots or commas to appear in recognition results.
ASPOSE_OCR_DETECT_CURVED_TEXT	4	Detect image areas containing cylindrical curvature distortions. Such areas may not be recognized accurately or not recognized at all.
ASPOSE_OCR_DETECT_BLURED_IMAGE	8	Detect blurry images, the recognition accuracy of which may be insufficient without preprocessing. Unlike other defect detection algorithms, this one cannot detect individual areas of the image. If it detects blurry text, the entire image area is returned.
ASPOSE_OCR_DETECT_GLARES	16	Detect highlight areas in an image caused by uneven lighting, such as spot lights or flash. Such areas usually have low contrast, which can negatively affect recognition accuracy or even lead to some texts not being recognized.
ASPOSE_OCR_DETECT_EXTRA_BOLD_TEXT	32	Detect very thick characters on an image. Such characters may be recognized incorrectly.
ASPOSE_OCR_DETECT_ALL	9999	Detect all supported image defects.

Example

The following code example shows how to fine-tune recognition:

// Provide the image
string file = "source.png";
AsposeOCRInput source;
source.url = file.c_str();
std::vector<AsposeOCRInput> content = { source };
// Fine-tune recognition
RecognitionSettings settings;
settings.language_alphabet = language::ukr;
settings.auto_contrast = true;
settings.upscale_small_font = true;
// Extract text from the image
auto result = asposeocr_recognize(content.data(), content.size(), settings);
// Output the recognized text
wchar_t* buffer = asposeocr_serialize_result(result, buffer_size, export_format::text);
std::wcout << std::wstring(buffer) << std::endl;
// Release the resources
asposeocr_free_result(result);