img_seg.py

import cv2
import numpy as np
import sys

def getAreaOfFood(img1):
	img = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY
	img_filt = cv2.medianBlur( img, 5)
	img_th = cv2.adaptiveThreshold(img_filt,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,11,2)
	contours, hierarchy = cv2.findContours(img_th, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)

	# find contours. sort. and find the biggest contour. the biggest contour corresponds to the plate and fruit.
	mask = np.zeros(img.shape, np.uint8)
	largest_areas = sorted(contours, key=cv2.contourArea)
	cv2.drawContours(mask, [largest_areas[-1]], 0, (255,255,255,255), -1)
	img_bigcontour = cv2.bitwise_and(img1,img1,mask = mask)

	# convert to hsv. otsu threshold in s to remove plate
	hsv_img = cv2.cvtColor(img_bigcontour, cv2.COLOR_BGR2HSV)
	h,s,v = cv2.split(hsv_img)
	mask_plate = cv2.inRange(hsv_img, np.array([0,0,100]), np.array([255,90,255]))
	mask_not_plate = cv2.bitwise_not(mask_plate)
	fruit_skin = cv2.bitwise_and(img_bigcontour,img_bigcontour,mask = mask_not_plate)

	#convert to hsv to detect and remove skin pixels
	hsv_img = cv2.cvtColor(fruit_skin, cv2.COLOR_BGR2HSV)
	skin = cv2.inRange(hsv_img, np.array([0,10,60]), np.array([10,160,255])) #Scalar(0, 10, 60), Scalar(20, 150, 255)
	not_skin = cv2.bitwise_not(skin); #invert skin and black
	fruit = cv2.bitwise_and(fruit_skin,fruit_skin,mask = not_skin) #get only fruit pixels

	fruit_bw = cv2.cvtColor(fruit, cv2.COLOR_BGR2GRAY)
	fruit_bin = cv2.inRange(fruit_bw, 10, 255) #binary of fruit

	#erode before finding contours
	kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(3,3))
	erode_fruit = cv2.erode(fruit_bin,kernel,iterations = 1)

	#find largest contour since that will be the fruit
	img_th = cv2.adaptiveThreshold(erode_fruit,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,11,2)
	contours, hierarchy = cv2.findContours(img_th, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
	mask_fruit = np.zeros(fruit_bin.shape, np.uint8)
	largest_areas = sorted(contours, key=cv2.contourArea)
	cv2.drawContours(mask_fruit, [largest_areas[-2]], 0, (255,255,255), -1)
	#dilate now
	kernel2 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(13,13))
	mask_fruit2 = cv2.dilate(mask_fruit,kernel2,iterations = 1)
	res = cv2.bitwise_and(fruit_bin,fruit_bin,mask = mask_fruit2)
	fruit_final = cv2.bitwise_and(img1,img1,mask = mask_fruit2)
	#find area of fruit
	img_th = cv2.adaptiveThreshold(mask_fruit2,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,11,2)
	contours, hierarchy = cv2.findContours(img_th, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
	largest_areas = sorted(contours, key=cv2.contourArea)
	fruit_contour = largest_areas[-2]
	fruit_area = cv2.contourArea(fruit_contour)

	
	#finding the area of skin. find area of biggest contour
	skin2 = skin - mask_fruit2
	#erode before finding contours
	kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(3,3))
	skin_e = cv2.erode(skin2,kernel,iterations = 1)
	img_th = cv2.adaptiveThreshold(skin_e,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,11,2)
	contours, hierarchy = cv2.findContours(img_th, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
	mask_skin = np.zeros(skin.shape, np.uint8)
	largest_areas = sorted(contours, key=cv2.contourArea)
	cv2.drawContours(mask_skin, [largest_areas[-2]], 0, (255,255,255), -1)

	skin_rect = cv2.minAreaRect(largest_areas[-2])
	box = cv2.cv.BoxPoints(skin_rect)
	box = np.int0(box)
	mask_skin2 = np.zeros(skin.shape, np.uint8)
	cv2.drawContours(mask_skin2,[box],0,(255,255,255), -1)

	pix_height = max(skin_rect[1])
	pix_to_cm_multiplier = 5.0/pix_height
	skin_area = cv2.contourArea(box)


	return fruit_area, mask_fruit2, fruit_final, skin_area, fruit_contour, pix_to_cm_multiplier


if __name__ == '__main__':
	img1 = cv2.imread(sys.argv[1])
	area, bin_fruit, img_fruit, skin_area, fruit_contour, pix_to_cm_multiplier = getAreaOfFood(img1)

	cv2.waitKey()
	cv2.destroyAllWindows()