OCR of the screenshot and blurring by regexp, Delphi

We need to create a screen recorder on Delphi 10.4 with 1sec fps what will be blurring sensitive text on the image by regular mask. As a result you will create a program that saves bmp files (screenshots) every 1sec in the program's subfolder. E.g it may work as a timer with 1sec with this code: function GetScreenShotBMP(Quality: TPixelFormat = pf32bit; A8bitLimit: Cardinal = 0): TBitmap; const CAPTUREBLT = $40000000; var DesktopCanvas: TCanvas; DC: HDC; Left, Top: Integer; begin Result := [login to view URL]; try DC := GetDC(0); [login to view URL] := Quality; try if (DC = 0) then Exit; [login to view URL] := [login to view URL]; [login to view URL] := [login to view URL]; Left := [login to view URL]; Top := [login to view URL]; if A8bitLimit > 0 then //use 8bit to avoid out of memory on many monitors begin if ([login to view URL] * [login to view URL]) > A8bitLimit then [login to view URL] := pf8bit; end; DesktopCanvas := [login to view URL]; try [login to view URL] := DC; BitBlt( [login to view URL],0,0,[login to view URL],[login to view URL], [login to view URL],Left,Top,SRCCOPY { CAPTUREBLT} ); //sleep(1000); finally [login to view URL]; end; finally if (DC <> 0) then ReleaseDC(0, DC); end; except [login to view URL]; Result := nil; end; end; The problem is, that the blurring code should work fast, much faster than 1sec. We need it to markup text or blur it for 0.1-0.3sec on popular cpu like i5. It means that your algorithm should detect new text on updated regions of the image only to except from OCR of the same parts.. Also, the fact that OS shows only printed fonts (not handmade) and their quality is 100% should also help to OCR faster... also, another possible way is to work from IAccessible interface... Our tests show that full screen image text recognition in tesseract takes 3-7sec. Also, if we have no screen update, we do not need to search for new text regions at all... The basic test is to blur credit card numbers or telephone numbers.