class DatasetCreator:
ESC = 27
DEL = 127
RET = 13
KEY_UP = 0
KEY_DOWN = 1
EXT = ".bmp"
FINAL_SIZE = 50
def __init__(self, prefix, dryRun):
h, w = 300, 300
img = np.zeros((h, w, 3), np.uint8)
img[:,:] = (255, 255, 255)
self.brush = Brush("Brush")
self.sketcher = Sketcher("Dataset Creator", img, self.brush)
self.__folders = OCR.generateFolderList(OCR.DIGITS | OCR.LETTERS | OCR.SYMBOLS)
self.__prefix = prefix
self.__lastFile = None
self.__dryRun = dryRun
def run(self):
while True:
k = cv2.waitKey(0) & 0xFF
if k == self.ESC:
break
elif k == self.RET:
self.sketcher.reset()
self.sketcher.show()
elif k == self.KEY_UP:
self.brush.brushSize += 1
elif k == self.KEY_DOWN:
self.brush.brushSize -= 1
elif k == self.DEL and not self.__dryRun:
if self.__lastFile:
os.remove(self.__lastFile)
self.__lastFile = None
print path + " Removed"
elif chr(k) in self.__folders:
folder = self.__folders[chr(k)]
path = common.generateFilename(folder, self.__prefix, self.EXT)
item = self.sketcher.sketch.copy()
item = cv2.resize(item, (self.FINAL_SIZE, self.FINAL_SIZE))
cv2.imshow("Generated Dataset item", item)
self.sketcher.reset()
self.__lastFile = path
if not self.__dryRun:
cv2.imwrite(path, item)
print path + " Written"
else:
print path + " Preview"
class TestOCR:
ESC = 27
KEY_UP = 0
KEY_DOWN = 1
def __init__(self, model, filenames, flags, pattern):
self.ocr = OCR()
self.ocr.loadModel(model, OCR.MODEL_ANN, flags)
if filenames:
counter = {}
for filename in filenames:
ch = self.ocr.charFromFile(filename)
if pattern:
label = self.__exec(pattern, filename)
if label in counter:
predicted, total = counter[label]
counter[label] = (predicted + int(label == ch), total + 1)
else:
counter[label] = (int(label == ch), 1)
print "%s in %s" % (ch, filename)
if pattern:
self.__analyze(counter)
else:
h, w = 200, 200
img = np.zeros((h, w, 3), np.uint8)
img[:,:] = (255, 255, 255)
self.brush = Brush("Brush")
self.sketcher = Sketcher('Test OCR', img, self.brush)
def __analyze(self, counter):
print ""
totalPercent = []
for label, (predicted, total) in sorted(counter.iteritems()):
percent = int(float(predicted) / total * 100)
totalPercent.append(percent)
print "{0}\t{1} / {2} - {3} %".format(label, predicted, total, percent)
print "\nTotal recognized : {0:.1f} %\n".format(np.mean(totalPercent))
def __exec(self, file, arg):
out = subprocess.Popen([file, arg], stdout=subprocess.PIPE).communicate()[0]
return out
def __displayResponse(self, ch):
response = np.zeros((75, 75, 3), np.uint8)
response[:,:] = (255, 255, 255)
cv2.putText(response, ch, (15, 35), cv2.FONT_HERSHEY_PLAIN, 3.0, (0, 0, 0), 1)
cv2.imshow("Response", response)
def run(self):
while True:
k = cv2.waitKey(0) & 0xFF
if k == self.ESC:
break
elif k == ord('r'):
self.sketcher.reset()
self.sketcher.show()
elif k == ord(' '):
ch = self.ocr.charFromImage(self.sketcher.sketch)
self.sketcher.show()
self.sketcher.reset()
self.__displayResponse(ch)
print "OCR : %s" % ch
elif k == self.KEY_UP:
self.brush.brushSize += 1
elif k == self.KEY_DOWN:
self.brush.brushSize -= 1