count.py

#from cv2 import cv
import cv2
import random
import sys
import numpy as np

print cv2.__version__

DICE_SIZE = 16
BLUR_FACTOR = 5
RED_LOW_THRESHOLD = 209
MIN_PIP_AREA = 10

def resizeRect(rect, sizeFactor):
	return (rect[0], (rect[1][0] + sizeFactor,rect[1][1] + sizeFactor), rect[2])
	

#img = cv2.imread("dice3.jpg")
img = cv2.imread("dice-real.jpg")
#img = cv2.imread("dice-real-2.jpg")

### Threshold image

blurred = cv2.medianBlur(img,BLUR_FACTOR)

blue = cv2.split(blurred)[0]
green = cv2.split(blurred)[1]
red = cv2.split(blurred)[2]

# Fetch the dice contours using red threshold. invert.
diceblocks = cv2.threshold(red, RED_LOW_THRESHOLD, 255, 1) # 185 --> 235
invdiceblocks = 255 - diceblocks[1]
##cv2.imshow("diceblocks",invdiceblocks)

# do a distance transform and normalize that so we can visualize and threshold it 
pyramids = cv2.distanceTransform(invdiceblocks, 2, 3)
cv2.normalize(pyramids, pyramids, 0, 1.2, cv2.NORM_MINMAX)

# obtain markers for the watershed algorithm by thresholding
markers = cv2.threshold(pyramids, 0.8, 1, 0)[1] 
##cv2.imshow("markers",markers)

# dilate the dice markers with a DICE_SIZE px element to capture all pips in the contours
#newImg = cv2.dilate(markers, cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(DICE_SIZE, DICE_SIZE)))

# convert the numpy matrix from float [0..1] to int [0..255]
bwImg = cv2.convertScaleAbs(markers * 255)

# capture those contours!
_, pyramids, hierarchy = cv2.findContours(bwImg.copy(),cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
print str(len(pyramids)) + " dice."

# fit a rotated rectangle on the distance transformed pyramid
for pyramid in pyramids:
	rect = cv2.minAreaRect(pyramid)
	# resize it to a certain dice size
	rect = resizeRect(rect, DICE_SIZE)

	floatBox = cv2.boxPoints(rect)
	intBox = np.int0(floatBox)
	bwImg = cv2.drawContours(bwImg,[intBox],0,(255,0,0),-1)

	pts1 = floatBox
	a,b,c,d = cv2.boundingRect(intBox)
	pts2 = np.float32([[a,b],[a+c,b],[a,b+d],[a+c,b+d]])

	M = cv2.getPerspectiveTransform(pts1,pts2)
	dst = cv2.warpPerspective(bwImg,M,pts2.shape)
	
# capture those large contours!
_, contours, hierarchy = cv2.findContours(bwImg.copy(),cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
	
# filter out the pips, and then cut out those using the contour areas
pips = 255 - cv2.threshold(cv2.cvtColor(blurred, cv2.COLOR_RGB2GRAY), 200, 255, 1)[1]
onlypips = cv2.bitwise_and(bwImg,pips)
#cv2.imshow("onlypips", onlypips)
dice = cv2.cvtColor(onlypips, cv2.COLOR_GRAY2RGB)

dice_results = [0,0,0,0,0,0]
wrongdice = 0

# look at each dice face and determine number of pips
for contour in contours:	
	pips = 0

	# get the coordinates to cut out the dice face from the image
	# with only the pips
	rect = cv2.minAreaRect(contour)
	floatBox = cv2.boxPoints(rect)
	intBox = np.int0(floatBox)
	a,b,c,d = cv2.boundingRect(intBox)
	
	# cut out the dice face
	subimage = onlypips[b:b+d,a:a+c]
	# count the number of contours
	_,pip_contours, subhierarchy = cv2.findContours(subimage.copy(),cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
	for pip in pip_contours:
		# count pips only if they are of a certain size
		if cv2.contourArea(pip) >= MIN_PIP_AREA:
			pips = pips + 1
		
	# log erroneous dice
	if pips > 6 or pips == 0:
		wrongdice = wrongdice + 1
		print pips
	else:
		dice_results[pips - 1] = dice_results[pips - 1] + 1
		cv2.putText(dice,str(pips),(a,b-5),0,1,(0,0,255))
	
# print out the results
print dice_results
print str(wrongdice) + " erroneous objects found."

cv2.drawContours(dice,contours,-1,(255,255,0),1)
cv2.imshow('Dice', dice)
cv2.imshow('Original',img)
#cv2.cvtColor(blurred, cv2.COLOR_RGB2GRAY))

def doCallbackTest(value):
	# newImg = cv2.adaptiveThreshold(newImg,255,1,0,13,value)
	
	
	tmpImg = red.copy()
	newImg = 255 - cv2.threshold(tmpImg, value, 255, 1)[1] #cv2.threshold(src, thresh, maxval, type
	#cv2.drawContours(tmpImg,contours,-1,(0,255,0),1)
	
	cv2.imshow('Dice',newImg)
	#cv2.imshow('Original',tmpImg)

lowThreshold = 1
max_lowThreshold = 255
#cv2.namedWindow("Dice", 0)
#cv2.createTrackbar('Value','Dice',lowThreshold, max_lowThreshold, doCallbackTest)

#doCallback(lowThreshold)

cv2.waitKey(0)
cv2.destroyAllWindows()