Generating ClearType subpixel positioned and all other Glyphs

Question

Generating ClearType subpixel positioned and all other Glyphs

polcott 20

I just provided a purely hypothetical guess of an answer to show exactly what an answer looks like.

I need to be able to generate and capture all of the rasterised images of individual font instances such as: {Times New Roman, Bold, Italic, 8-Point @96 dpi} That any Microsoft desktop application would generate I need this to train my DFA OCR.

Microsoft has several different ways that it can generate bitmap images of glyphs. The following example was captured from a browser window seems to use the latest ClearType that has subpixel positioning. four_glyphs

I need to be able to generate and capture all of the different bitmap images of glyphs all of the ways that Microsoft can do this for every Microsoft application on a single desktop computer. WordPad only has one glyph bitmap per character.

WFP can generate the latest ClearType glyphs. DirectX can also do this, yet I do not know that the resulting bitmaps are identical. I don't know how TrueType and OpenType fit into this picture.

I also need to capture the bitmaps of these glyphs from the Window. I am unsure of which Microsoft technologies are best for this. WPF is my current starting point.

Can I generate all of the bitmaps of all of the font instances for a every application used on a specific machine using WPF, or do I need separate programs that also use GDI, GDI+, and DirectX?

sam 0 Reputation points

2023-07-12T21:32:54.04+00:00

In WPF (Windows Presentation Foundation), you can generate bitmaps of font instances for applications that use WPF for rendering. However, to generate bitmaps for applications that use GDI (Graphics Device Interface), GDI+, or DirectX for rendering, you would need separate programs that utilize those specific technologies.

WPF has built-in support for rendering vector-based fonts using the glyph rendering capabilities of DirectWrite.Longboard Tricks You can use WPF to generate bitmap representations of fonts within WPF applications by rendering text onto a RenderTargetBitmap object. This allows you to capture the visual appearance of the rendered text as a bitmap image.

However, if you want to generate bitmaps for applications that rely on GDI or GDI+ for rendering text, you would need to use GDI or GDI+ specific APIs and techniques. Similarly, for applications that utilize DirectX for rendering, you would need to utilize DirectX APIs to generate the bitmaps.

Each technology (WPF, GDI, GDI+, DirectX) has its own set of APIs and rendering mechanisms, so generating bitmaps for applications that use different rendering technologies may require separate programs that use the corresponding APIs. It is not possible to directly generate bitmaps for all font instances across different rendering technologies from a single program using just WPF.

2 answers

Your answer

sam 0 Reputation points

2023-07-12T21:32:54.04+00:00

In WPF (Windows Presentation Foundation), you can generate bitmaps of font instances for applications that use WPF for rendering. However, to generate bitmaps for applications that use GDI (Graphics Device Interface), GDI+, or DirectX for rendering, you would need separate programs that utilize those specific technologies.

WPF has built-in support for rendering vector-based fonts using the glyph rendering capabilities of DirectWrite.Longboard Tricks You can use WPF to generate bitmap representations of fonts within WPF applications by rendering text onto a RenderTargetBitmap object. This allows you to capture the visual appearance of the rendered text as a bitmap image.

However, if you want to generate bitmaps for applications that rely on GDI or GDI+ for rendering text, you would need to use GDI or GDI+ specific APIs and techniques. Similarly, for applications that utilize DirectX for rendering, you would need to utilize DirectX APIs to generate the bitmaps.

Each technology (WPF, GDI, GDI+, DirectX) has its own set of APIs and rendering mechanisms, so generating bitmaps for applications that use different rendering technologies may require separate programs that use the corresponding APIs. It is not possible to directly generate bitmaps for all font instances across different rendering technologies from a single program using just WPF.

Answer 1

Bruce (SqlWork.com) 78,006 Volunteer Moderator

a Glyph in ClearType is a single font character. It is not a raster image but a small TrueType program. TrueType defines glyphs as a series of quadratic Bézier splines and hints. it's a small push/pop machine language.

https://learn.microsoft.com/en-us/typography/truetype/fixing-rasterization-issues
https://learn.microsoft.com/en-us/typography/truetype/hinting-tutorial/hinting-and-truetype-instructions

ClearType calls TrueType to produce a raster image that ClearType then performs a filter on. This is done at runtime to produce a raster image based on the font size and screen resolution.

you can use the TrueType tool to produce a raster image of a font character in varies font sizes and screen resolutions.

https://learn.microsoft.com/en-us/typography/tools/vtt/

note: with modern hi-res displays, ClearType is fairly obsolete and been replaced with OpenType (an extended truetype format) which supports larger font families.

polcott 20 Reputation points

2023-07-10T22:28:33.3533333+00:00

I have captured Glyphs that Microsoft Edge generates on Windows 10 and there are four distinct glyphs per character. Original ClearType has only one. PDF files (that do not use MS technologies) have three distinct glyphs per character.

I need to capture the bitmap images of every rasterized font every way that Microsoft can do this.
Viorel 122.6K Reputation points

2023-07-11T09:04:54.3+00:00

Are you interested in results produced by different options of Graphics.TextRenderingHint?
Bruce (SqlWork.com) 78,006 Reputation points Volunteer Moderator

2023-07-11T16:03:15.6233333+00:00

the number of glyph raster images will depend number of hints defined for the font glyph. this can vary character by character.
polcott 20 Reputation points

2023-07-11T19:31:53.7866667+00:00

I need to be able to generate and capture all of the rasterised images of individual font instances (for example (Times New Roman, Bold, Italic, 8-Point) using every Microsoft technology that is currently available. I need this to train my DFA OCR.
Bruce (SqlWork.com) 78,006 Reputation points Volunteer Moderator

2023-07-11T21:15:44.5333333+00:00

most font hints are not required after a certain size/resolution.

for example your sample are for a font rendered at 12 pixels. on my monitor which is 218 pixels per inch, the above glyphs would be .055 inch tall. probably not scannable by your OCR.

fonts are type measured in points (72/inch). so on my monitor a 8 point font is 24 pixels tall, twice the resolution of your sample glyphs.

the raster with hints is an algorithm. the hints have more effect an smaller raster images. on an old 72 dpi lcd monitor an 8 point font was only 8 pixels tall. anti-aliasing by the sub pixels made a big difference.

so you need to sample your 8 point font at different resolutions. common monitor resolution sizes are, 360, 480, 800, 1024, 1280, 1360, 1366, 1440, 1536, 1600, 1680, 1920, 2048, 2560, 3440, 3640. the dpi will be different depending on common monitor sizes (12, 13, 15, 17, 19, 20, 27, 32) and font scaling (normal, small, large).

this gives you a lot of variations. for 8 point point, if you rastered at 8-24 dpi you would probably get most of the variants. if you can not control dpi, pick a point size that is gives glyph about 8 pixels tall for the tallest glyph in the fontset (probably j).

say it's 6 point. then render from 6 point to 24 point. this should give you most variants. you could probably get away with 6, 8, 10, 12, 18, 24

of course, there is the TrueType, ClearType, OpenType variants of the font.

note: this is all about the screen raster images. It's a different story for printers. Printers tend to need hints for fonts < 12 or 18 point, because they typically start at 300 dpi).
polcott 20 Reputation points

2023-07-11T22:20:25.7766667+00:00

I am referring to the different technologies for generating the bitmap of a FontInstance on a single machine. For example I know that ClearType alone does this at least three different ways with a different number of bitmaps per character. I know that both WPF and DirectX both generate Sub-pixel Positioning and Y-Direction Antialiasing, yet do not know if the generated bitmaps are identical on the same machine for the same FontInstance. I also don't know how a user would select TrueType, ClearType, and OpenType variations of a font. ----------------------------------------------------------------------------------------------------
Bruce (SqlWork.com) 78,006 Reputation points Volunteer Moderator

2023-07-11T23:12:06.4233333+00:00

on a single machine, any font technology with hints will produce different bitmaps at different font sizes. the greatest variance will be at small font sizes.

subpixel is a ClearType feature (and only for lcd screens, should be disabled for an oled display). Also ClearType is poor choice for screens that can rotate (like a tablet), as they are only optimized for one direction.

the font file determines the technology used to rasterize a font. so the font chosen by the user determines the technology.

if you include browsers, they can dynamically download a font, and there are several formats supported by the browser.

OpenType supports TrueType and Postscript Type 1 font data.

With a drawing package, the code can render the font in a particular font size, and then perform a matrix operation that changes the rotation and size. Obviously sub pixels don't work well on a rotated glyph which is why you might not want to use a ClearType font.
polcott 20 Reputation points

2023-07-11T23:28:08.0233333+00:00

I need to be able to write a computer program that generates the bitmaps for a particular font instance such as {Times New Roman, Bold, Italic, 8-point} every different way that is possible on that machine. For example on my own machine I found out that this has one bitmap per character in WordPad and four bitmaps per character in Microsoft Edge and Google Chrome with different colored bitmaps on Chrome. I am only dealing with desktops and laptops.

So to make my requirements concrete I need to generate the exact same bitmaps that Microsoft Edge, Google Chrome, WordPad and every other app that can be executed on my desktop would generate for a specific font instance. I need to do this without running these apps because the amount of data is too massive 900,000 lines of text as bitmaps per per font instance per application.

Can this be done entirely in WPF? or does this have to be done by also using GDI, GDI+ and DirectX?

Answer 2

polcott 20

I am providing this hypothetical answer to show exactly what kind of an answer I am looking for. It is not fact based it is merely a guess of what the facts could be.

Windows Presentation Foundation (WPF) can generate the exact same glyph bitmap images that every other Windows technology {GDI, GDI+, DiredtX} can generate. This will work consistently for TrueType, OpenType and any version of ClearType.

This will allow generating and capturing the bitmaps of 900,000 lines of text that will exactly match the bitmaps generated by every Microsoft application by any of the Microsoft technologies and any font technology when done on the same desktop computer/monitor configuration. In some of these cases the rasterising parameters may need to be adjusted.

Bruce (SqlWork.com) 78,006 Reputation points Volunteer Moderator

2023-07-12T16:04:31.13+00:00

I'm not sure. for example GDI+ apps can define their own font aliasing. Also for example, with a dynamic font, the location, kern, etc of the character can change the raster. There are also lots of complaints of different Microsoft technologies (WPF, GDI, GDI+, DirectX) not having consistent font rendering.

I'm not sure your OCR scanning software, but it should NOT depend on exact bitmap matches.
polcott 20 Reputation points

2023-07-12T16:14:38.91+00:00

Without depending on exact bitmap matches my technology becomes infeasible because it is based on a deterministic finite automaton (DFA) and not the stochastic process that conventional OCR uses. Its advantage is that for unambiguous font instances it is 100% accurate. It is also very much faster than conventional OCR. It is used for data interchange between GUI applications having no other interface for data interchange. ---
Bruce (SqlWork.com) 78,006 Reputation points Volunteer Moderator

2023-07-12T19:48:26.84+00:00

then its not likely to be reliable. you would be hard pressed to get the complete raster set for a single font. for each char in a font, there are dozens of font sizes and for each size there a different raster images based on screen location (x,y) and enabled hints. also most apps allow some sort of zoom, which adds a new set of raster images.

also the anti-aliasing will vary based on the font color and the background color.

also advanced text apps that support kerning will cause overlaps of the raster images. for example a kerned "ev", "Ko", or "av", "fi", "ffl", etc. (though some fonts will have kerned ligatures)
polcott 20 Reputation points

2023-07-12T20:05:07.6266667+00:00

As I have been repeatedly stating I am only concerned about a single font size on a single machine. I have empirically validated that there are four bitmaps per character for the current subpixel positioning with Y-Direction Antialiasing. The font background and foreground colors will also be fixed constants. Per customer I only need a single fixed contant font instance generated by a specific Microsoft based application. This will be only generated by a single Microsoft technology. My question here has always been do I need four different programs to accomplish this (GDI, GDI+, DirectX, and WPF) or can I do it all in WPF? (I have done a relatively terrible job of asking this question, sorry).
Bruce (SqlWork.com) 78,006 Reputation points Volunteer Moderator

2023-07-12T21:58:07.7266667+00:00

GDI supports three font types. bitmap, vector and TrueType

GDI+ uses bitmap fonts with its own anti-aliasing

DirectX uses bitmap fonts (generated by the developer from truetype fonts) unless DirectWrite() is used. DirectWrite() uses ClearType but allows some control of hints.

you will probably need to produce samples with each.

note: I would not expect your approach to be as reliable as other OCR approaches.
polcott 20 Reputation points

2023-07-12T22:10:00.2833333+00:00

It has proven to be much more reliable than other approaches. On very tiny {Times New Roman, Bold, Italic, 8-point, Black on White} it scored 100% accuracy the next most accurate OCR in the world only got 25% accuracy and was 872-fold slower.
Bruce (SqlWork.com) 78,006 Reputation points Volunteer Moderator

2023-07-12T22:53:08.7766667+00:00

on my Mac where select text from an image is builtin, if I screenshot (make a png image) of 8 point text. I can open the image, and use the builtin OCR support to copy the text from the image, no problem. It is actually almost too small for me to read.

I also tried 6 point, and the OCR could handle fine, but I needed to zoom the png image to be able to select the text.

note: selecting and copying text from an image is as fast as normal text.
polcott 20 Reputation points

2023-07-12T23:03:56.8266667+00:00

No stochastic process can possibly beat the exact match of my DFA based technology. I have worked on this for 20 years and have two patents. Try the same test on a script based font. If it does as well I would suspect that the system is cheating. Also 8-point on a Mac is probably much more than 10 pixels tall. OCRtest.gif

This is the text sample that my OCR scored 100% on it was 872-fold faster than Adobe OCR:

(0.047 seconds compared to 41 seconds)---

**
polcott 20 Reputation points

2023-07-12T23:22:43.48+00:00
Bruce (SqlWork.com) 78,006 Reputation points Volunteer Moderator

2023-07-13T01:42:50.36+00:00

less than a second to select. you are correct, trouble with some of the letters. copy paste:

MCFC has a class called eDialeg which is a CWnd desived class
that is specilically wed lex the creatien and display of dialeg cones on screen - coin modar and moderess diaccas are supported. Modeters diatcas are created using Create and you have la destroy them on your ond wunig Destreylindow and they cahave wot tice any nermat meddless windem. sledal dialogs ave created and shown using the sermidate Dastedal methed el the Diates dass. And yeu dese a medal dialeg bu calling EndDialog) ex alternatindy by catting OnG) as OnCancel() beth el which internally call EndDialeg(). The

Dialog. EndDialog mdthed will call the EndDialeg Win32 (33 Kundion which is defined in wer32 dll. EndDialog (the (.93

fundion) will not immediately deve the dialog, ather it will od destrou the diaics window and cnatte the pasent wundoms sa an da good; everytrung seems so peacelut and severe, and the lillle

------ dropped line ----

curds in the tree acvoss the ward are tiesting in a beautiful

-- dropped word
polcott 20 Reputation points

2023-07-13T03:49:46.9233333+00:00

100% accuracy in 0.047 seconds

MFC has a class called CDialog which is a CWnd derived class that is specifically used for the creation and display of dialog boxes on screen - both modal and modeless dialogs are supported. Modeless dialogs are created using Create() and you have to destroy them on your own using DestroyWindow() and they behave just like any normal modeless window. Modal dialogs are created and shown using the formidable DoModal() method of the CDialog class. And you close a modal dialog by calling EndDialog() or alternatively by calling OnOK() or OnCancel() both of which internally call EndDialog(). The CDialog::EndDialog method will call the EndDialog Win32 API function which is defined in user32.dll. EndDialog (the API function) will not immediately close the dialog, rather it will set a flag which will instruct the message queue to exit the loop, destroy the dialog window and enable the parent window. So far so good; everything seems so peaceful and serene, and the little birds in the tree across the yard are tweeting in a beautiful voice.

Share via

Generating ClearType subpixel positioned and all other Glyphs

2 answers

Your answer