TIEVisionNNet.detectTexts

Declaration

function detectTexts(model: int32_t; image: TIEVisionImage; out confidences: TIEVisionVectorDouble; out detections: TIEVisionVectorRotatedRect): bool32; safecall;

Description

This method takes an input image and returns the detected texts along with their confidences and bounding boxes.
Returns True on success (both input and model valid).

Parameter	Description
model	Model type. 0 = EAST, 1 = DB
image	Input image
confidences	Output array of confidences per text block detected
detections	Output array rotated rectangles per text block detected

Note: This method only detects the location of text. To "Recognize" the text, use OCR

Example

// Detect all text blocks inside TImageEnView using DB_IC15_resnet18.onnx pretrained model
var
  nnet: TIEVisionNNet;
  rrects: TIEVisionVectorRotatedRect;
  confidences: TIEVisionVectorDouble;
  i: integer;
  r: TIEVisionRect;
begin
  OCR := IEVisionLib.createOCR(IEOCRLanguageList[OCR_English_language].Code);
  nnet := IEVisionLib.createNNet('models\DB\DB_IC15_resnet18.onnx');
  nnet.setInputSize(1280, 736);
  nnet.setInputScale(1.0/255);
  nnet.detectTexts(1, ImageEnView1.IEBitmap.GetIEVisionImage(), confidences, rrects);
  for i := 0 to rrects.size() - 1 do
  begin
    r := IEVisionRect( rrects.getRect(i) );
    // Note: With the East model in can be useful to inflate the text boxes to improve detection
    s := OCR.recognize( ImageEnView1.IEBitmap.GetIEVisionImage(), r ).c_str();
    Memo1.Lines.Add( s );
  end;
end;

TIEVisionNNet.detectTexts