def getText(image):
chatfont = FontData("w:/app/poker/ps-chat.fontd")
for top,base,bottom in scrape.guess_lines(image):
def hasInk(a,b):
return image.getpixel((a,b+top+1)) == 0
yield scrape.getstring(w, bottom-top, chatfont, hasInk,
)#train=True)
def glyphs(img, font, cutoff=1, train=False):
"""
This is slightly more "advanced" OCR that I used
when trying to re-assemble joe's book
It attempts to deal with anti-aliased text.
However, the method it uses is rather silly.
generates (x,y, width, height, char) tuples for each glyph in an image.
glyphs may contain several characters because this system cannot
compensate for ligatures/kerning. This doesn't really
work too well yet.
"""
for top, baseLine, bottom in guess_lines(img.convert("1")):
scanH = top - bottom # 16 #top-bottom
fakeTop = top # baseLine - 14
fakeBot = bottom # baseLine + 3
def hasInk(a, b):
return img.getpixel((a, b + fakeTop)) <= cutoff # re-alias for clarity
w, h = img.size
x = 0
for charW, bmp in scan_line(w, scanH, hasInk):
if bmp > 0:
icon = iconify(img.copy().crop((x, top, x + charW, top + scanH)))
code = icon2num(icon)
if code in font:
char = font[code]
elif train:
print bmp
print convert.glint_to_strings(bmp, scanH)
char = raw_input("what is it?")
font[code] = char
else:
char = code
yield x, top, charW, scanH, char
elif bmp == 0 and not train:
yield x, top, charW, scanH, " "
x += charW