findLines_notes

Using Hough transform:

Accuracy not great - need to figure out if the parameters need to be tweaked, or if it is better to try something else.

----------------------------------------
Using Leptonica:

Seems to be working better. Can clean up, e.g., threshold image before handing to Tesseract.
Can recover lines using Leptonica.

Vertical lines - look at Tesseract code.

linefind.cpp l.618

in GetLineMask

----------------------------------------
Other routines we might need from Leptonica:

pixAddPix
pixOr
pixXor

pixOpenBrick
pixCloseBrick

pixGetWpl l.724
pixConnComp l.738 (might be exactly what we need)

boxCreate (lept)
boxDestroy (lept)

FindLineVectors

pixClearInRect
pixSubtract
pixDilateBrick
pixSeedfillBinary
pixDistanceFunction
pixClipRectangle
pixCountPixels
pixZero
pixErodeBrick

Leptonica is writing to console - need to capture tprintf()

pixGetPixel expects x, y
matrix in R is written y, x - so code needs to be clear

----------------------------------------
Using test image

ThresholdtoBinary = inverts values, low values -> 1

Out of pixThresholdToBinary():
1 = white
0 = black

out of getLines(..., asLines = TRUE, invert = !asLines):
1 = black
0 = white

pixGetPixels:
1 = black
0 = white

pixWrite() + readPNG():
1 = white
0 = black

Appears that readPNG and pixGetPixels are in the same order - there does not appear to be any flipping.

in ConvertTo8, colormap = FALSE, might cause issues with the color map

----------------------------------------

Look at linefind.cpp, see what they are doing -

Pre-cleaning is difficult, avoid cmdline, avoid mixing computational models