Better fix infinite recursion + Better line detection.

With this change I am assuming text that we want to extract with this actually contains some text, and not just a row of _ . , and -

Author: slackydev

simpleocr.engine.pas

@@ -27,6 +27,10 @@ interface
   FONTSET_START = #32;
   FONTSET_END   = #126;
 
+const
+  ALPHA_NUMERIC_SYM = ['a'..'z', 'A'..'Z', '0'..'9','%','&','#','$','[',']','{','}','@','!','?'];
+
+
 type
   PFontCharacter = ^TFontCharacter;
   TFontCharacter = packed record
@@ -95,6 +99,17 @@ implementation
 uses
   graphtype, intfgraphics, graphics, math;
 
+
+function ContainsAlphaNumSym(text: string): Boolean; inline;
+var i: Int32;
+begin
+  Result := False;
+  for i:=1 to Length(text) do
+    if Text[i] in ALPHA_NUMERIC_SYM then
+      Exit(True);
+end;
+
+
 function TFontSet.GetCharacterPoints(const Character: Char): Integer;
 begin
   if (Character in [FONTSET_START..FONTSET_END]) then
@@ -678,15 +693,21 @@ function TSimpleOCR.RecognizeLines(Matrix: TIntegerMatrix; Filter: TOCRFilter; c
       begin
         // OCR the row and some extra columns
         Text := Self._RecognizeXY(Box(SearchBox.X1, SearchBox.Y1, SearchBox.X2, SearchBox.Y1 + (FFontSet.MaxHeight div 2)), FontSet.CharacterPoints[Filter.MinCharacterMatch], $FFFFFF, Hits, Bounds);
-        if (Text = '') or (Bounds.Y1 = LastBounds.Y1) then
+
+        if (Text = '') then
           Exit;
 
-        LastBounds := Bounds;
-        Result := Result + [Text];
-        TextBounds := TextBounds + [Bounds];
+        // Ensure that actual text was extracted, not just a symbol mess of short or small character symbols.
+        if ContainsAlphaNumSym(Text) then
+        begin
+          LastBounds := Bounds;
+          Result := Result + [Text];
+          TextBounds := TextBounds + [Bounds];
 
-        // Move down to the found text Bounds.Y2 (minus a little) so we don't recognize this again
-        SearchBox.Y1 := Bounds.Y2 - (FFontSet.MaxHeight div 4);
+          // Now we can confidently skip this search line by a jump, but we dont skip it fully in case of overlapping text
+          // So we divide the texts max glyph height by 4, and subtract that from the lower end of the found bounds.
+          SearchBox.Y1 := Max(SearchBox.Y1, Bounds.Y2 - (FFontSet.MaxHeight div 4));
+        end;
       end;
 
       SearchBox.Y1 += 1;

tester/tester.lpr

@@ -129,7 +129,7 @@ procedure Test_MultiLine3;
   Lines := SimpleOCR.RecognizeLines(LoadMatrix('images/multiline3.png'), Filter1, FONT_PLAIN_11);
   for I := 0 to High(Lines) do
     Lines[I] := StringReplace(Lines[I], 'I', 'l', [rfReplaceAll]);
-
+  WriteLn(Lines[1]);
   Assert(Length(Lines) = 5);
   Assert(Lines[0] = 'Leather Boots:');
   Assert(Lines[1] = 'Adamant Kiteshield:');
@@ -292,6 +292,7 @@ procedure Test_Threshold1;
     MinCharacterMatch: #0;
   );
 begin
+  WriteLn(SimpleOCR.Recognize(LoadMatrix('images/thresh.png'), Filter, FONT_BOLD_12));
   Assert(SimpleOCR.Recognize(LoadMatrix('images/thresh.png'), Filter, FONT_BOLD_12) = 'Showing items: hello');
 end;