LEADTOOLS OCR Advantage Engine Settings

Recognition module tradeoff between speed and accuracy. The default value is Balanced (1).

Possible values are:

true to share the image reference used to create the OCR page; otherwise, false. The default value is false.

Setting this value to true affects the IOcrPageCollection.AddPage(RasterImage, OcrProgressCallback) and IOcrPageCollection.InsertPage(int, RasterImage, OcrProgressCallback) methods.

When the value is false (the default), these methods make a copy of the image and use the copy to create the page. Calling IOcrPage.GetRasterImage(OcrPageType.Original) on such a page returns a null reference.

When the value is true, these methods use the same image reference to create the page. Calling IOcrPage.GetRasterImage(OcrPageType.Original) on such a page returns the original image reference.

IOcrAutoRecognizeManager temporarily sets the value of this setting to true while performing a recognition job.

true to modify the processing image after recognition; otherwise, false. The default value is true.

It is best to set the value of this setting to true if IOcrPage.Recognize is called only once per page.

IOcrAutoRecognizeManager temporarily sets the value of this setting to true while performing a recognition job.

Automatically detect the foreground and background colors of each character. The default value is true.

If this value is true, then the engine tries to automatically detect the colors of the zones when IOcrPage.AutoZone is called and sets the values in OcrZone.ForeColor and OcrZone.BackColor.

Automatically perform second image processing cleanup on the internal black and white image if the first pass did not provide satisfactory results. The default value is true.

The appropriate setting for Recognition.MaximumPageConventionalMemorySize depends on the system hardware configuration and the number of cores and application types being used. Change this setting if out-of-memory errors occurs when running your application.

The IOcrEngine supports loading RasterImage objects directly from disk files, streams or URLs, (as for example, the various methods in the IOcrPageCollection and IOcrAutoRecognizeManager classes do). The RasterImage loaded holds the original image and is useful only when saving graphics zones or image-over-text overlays. If the image is large, and one created using conventional memory, then a large amount of physical memory is used to hold this image in memory and is not available for other purposes such as auto-zoning or recognition. This is more noticeable in multi-threaded applications where loading several large images in conventional memory can cause out-of-memory errors, even when performing operations that normally would succeed.

Use "MaximumPageConventionalMemorySize" to set the maximum size of the image in memory allowed before the IOcrEngine automatically switches to use the disk memory feature of RasterImage (RasterMemoryFlags).

The "MaximumPageConventionalMemorySize" is in KBytes and its default value depends on the processor(s) being used. For x86 processors, the value is 42187 (42MBytes). For x64 processors, the value is calculated dynamically (1.7GBytes for each 8 cores, not exceeding the physical memory size). These values allow a typical OCR image (8.5 by 11 inches at 300 DPI and 32-bits per pixel) to be loaded in conventional memory. Anything significantly larger than that gets switched to use disk memory mode.

Different factors affect the performance of a particular setting and must be weighed. These include the following factors:

• The speed of the machine's hard drive - increases the penalty for using disk memory rather than conventional memory
• Load time - using disk memory consumes more time loading than using conventional memory
• Autozoning and Recognition - using disk memory improves the performance of autozoning, recognition and other operations because conventional memory is freed for image processing.

Gets or sets the maximum number of threads to use in recognition. The LEADTOOLS Advantage OCR engine provides support for recognizing document zones in separate threads. This can improve the performance of the IOcrPage.Recognize method.

The default value is 0, which instructs LEADTOOLS to use the system thread pool. The number of threads is calculated automatically. Setting the value to a number > 1 also turns multi-threading on, but the number of threads specified is ignored and calculated automatically instead.

To turn multi-threading off inside the IOcrPage.Recognize method and use a single thread, set the value to 1.

This setting influences how a non-black and white image, stored in the Engine, is converted to a black and white one. The default value is Default (0).

Possible values are:

The threshold to use when converting colored images to bitonal (black and white) in preparation for recognizing the text on the image. Conversion separates the text intensities from the background intensities. The default value is 185.

This is the equivalent of calling IntensityDetectCommand on the image with InColor equal to the detected foreground (text) color, OutColor equal to the detected background color, Channel set to Master, HighThreshold equal to 255, and LowThreshold equal to the value of this setting.

true to enable mobile image processing mode; otherwise, false. The default value is false.

By design, the OCR engine tries to upscale low resolution (DPI) images. However, most cameras in mobile devices take pictures with a low resolution (for example, 72 DPI) but a large pixel size. If the OCR engine upscales such images, a lot more memory than necessary is consumed. If you are using the OCR engine to process images from a mobile camera, set the mobile image processing mode to true.

true to downsample large images prior to recognition; otherwise, false. The default value is false.

Set the value to true to prevent the OCR engine from creating processing images (the image used for recognition) larger than 4000 by 4000 pixels (in order to preserve memory and resources). This value is ignored if the value of the MobileImagePreprocess setting is true.

true to use the zoning engine to exclude graphics areas from preprocessing calculations such as deskew and auto-rotate; otherwise, false. The default value is true.

Used by the IOcrPage.AutoPreprocess method to determine the minimum confidence percentage threshold to use when orienting pages. The default value is 26.

Specifies how the original image is modified when calling the IOcrPage.AutoPreprocess method. The default value is Deskew | Rotate | Invert (0x01 | 0x02 | 0x04 = 0x07).

Values can be OR'd together. Possible values are:

These options are useful when saving a document that has image over text (such as PDF documents). When saving a document that has image over text it may be preferable to overlay the original image without making any modifications that could affect the size. The only modification that should be allowed in such a case is "Rotate". The IOcrAutoRecognizeManager automatically sets the value to "Rotate" if the final document format supports image over text.

true to automatically detect and recognize inverted regions, otherwise; false. The default value is false.

Set this value to true to support OCRing an image that contains both black/white and white/black areas.

true to disable multi-threading when performing auto-zoning; otherwise, multi-threading is enabled. The default value is false.

Multi-threading enhances the performance of the auto-zoning algorithm. However, it can be undesirable if the OCR engine is hosted on a server.

true to crop each zone from the original image and recognize it by itself; otherwise, false. The default value is true.

Setting this value to true can improve the performance of the IOcrPage.Recognize method.

true to try to detect a separate rotation angle for each zone; otherwise, false. The default value is false.

Setting this value to false can increase the speed of the recognition engine.

Vertical zone detection mode. This works with Latin and Asian languages.

Possible values are:

These flags affect the way the IOcrPage.AutoZone method works. The default value is Detect Text | Detect Graphics | Detect Table | Detect Accurate Zones | Table Cells As Zones | Use Advanced Table Detection | Use Text Extractor | FavorGraphics (0x01 | 0x02 | 0x04 | 0x10 | 0x40 | 0x80 | 0x100 | 0x400).

Values can be OR'ed together. Possible values are:

true to perform a second, internal autozoning procedure on each text zone in order to generate more homogenous zones for recognition; otherwise, false. The default value is true.

Setting this value can improve the performance of the IOcrPage.Recognize method.

true to discard words with a low rating; otherwise, false. The default value is true.

This setting controls the output. If true, the engine checks the confidence of each word and prevents any having a low rating (lower than the LowWordConfidence value) from being included when saving the recognition results to any of the document formats supported by the LEADTOOLS toolkits.

true to discard zones with low ratings; otherwise, false. The default value is false.

This setting controls the output. If true, the engine checks all of the words/characters in a zone. If it determines that the overall confidence and type of characters constitute noise, then the recognition results for the entire zone is discarded.

Discard any word with a confidence value less than this value. the default value is 50.

This setting only takes effect when DiscardLowConfidenceWords is set to true.

Not used in this version of LEADTOOLS

The minimum width of a recognized character, in pixels. The default value is 6.

The minimum height of a recognized character, in pixels. The default value is 6.

Characters to exclude from the minimum pixel width and height rule. The default value is ".".

true to ignore recognized characters that have features similar to noise; otherwise, false. The default value is false.

true to discard all the results in the zone if the engine determines that all the characters recognized are noise; otherwise, false. The default value is false.

true to post-process any MICR zones by discarding all of the characters, numbers and symbols that do not belong to the MICR character set as well as performing basic validity data checking; otherwise, false. The default value is true.

true to enable Caps/Caps (CamelCase) font recognition enhancements; otherwise, false. The default value is false.

Enable or disable the detection of specific font properties. These flags affect the final generated document if the format supports fonts (such as the PDF or DOCX formats). The default value is Bold | Italic | Underline | SansSerif | Serif | Proportional | Superscript | Subscript | Strikeout (0x01 | 0x02 | 0x04 | 0x08 | 0x10 | 0x20 | 0x40 | 0x80 | 0x100 = 1FF).

Values can be OR-ed together. Possible values are:

true to enable font attributes recognition; otherwise, false. The default value is true.

Setting this value to false can improve the speed of the IOcrPage.Recognize method if the font attributes of the recognized characters is not required, for instance, if recognition is used to obtain raw text only and not to create a formatted output document.

true to optimize the recognition speed based on the format of the final document; otherwise, false. The default value is true.

As an example, the OCR engine does not recognize font attributes such as italic or bold if the final document format is Text.

The default orientation for the generated document if a page is blank or entirely graphics. The default value is None.

Possible values are:

Gets or sets the maximum number of spell checkers to use at the same time. The default value is the number of available dictionaries found in the system.

true to enable caching of frequent words, otherwise; false. The default value is true.

This affects grayscale or 24-bit color images. A black and white image is created in the Engine's memory. Image binarization applies an automatic adaptive thresholding algorithm.

This affects grayscale or 24-bit color images. A black and white image is created in the Engine's memory. Each pixel is compared to a dynamically-calculated threshold. If the pixel intensity is higher it is set to white; otherwise, it is set to black.

This affects grayscale or 24-bit color images. A black and white image is created in the Engine's memory. Thresholding with a user-defined threshold value is performed. Set the threshold with Recognition.Preprocess.BlackWhiteImageConversionThreshold).