python 2.7 - How can i find the pixel color range in an image that excludes outliers? -

i have code randomly generates 10 pixel locations in image dimensions. want take each pixel location , find gbr value , find largest g value, b value, r value , smallest g value b value , r value. found code can take color boundaries , create mask image. might want incorporate mean g, b , r value , compute standard deviation each , use eliminate outliers.

or instead of using code wrote generate 10 random pixels if there way color range inside of +- 1 standard deviation of mean color of image?

see code below:

    import cv2 import numpy np import random  name = "highway"  img = cv2.imread(name + ".jpg") inc = 10 n = 10 rpixel=[]  # grabs bottom center of image, keeps image ratio. larger inc smaller imaged grabed  def bottomcenter(inc,img):      y,x,z = img.shape     h = x/2     x1 = h - (x/inc)     x2 = h + (x/inc)     y1 = y - (y/inc)     y2 = y       bcsample = img[y1:y2, x1:x2]      return(bcsample)   # generates random selection of pixels  def randompixel(img,n):      y,x,z = img.shape      in range (n):         xrand = random.randrange(0,x)         yrand = random.randrange(0,y)         rpix = yrand,xrand         rpixel.append(rpix)     return(rpixel)    roadsample = bottomcenter(inc,img)  randompixel(roadsample,n)     in rpixel:      #px=b.g.r        px = roadsample[i]      b,g,r in px:          ub = max(b)         ug = max(g)         ur = max(r)         lb = min(b)         lg = min(g)         lr = min(r)  boundary = [([lb,lg,lr],[ub,ug,ur])]  print(boundary)  #for (lower, upper) in boundary:     # create numpy arrays boundaries #   lower = np.array(lower, dtype = "uint8") #   upper = np.array(upper, dtype = "uint8")  # find colors within specified boundaries , apply # mask #mask = cv2.inrange(img, lower, upper) #output = cv2.bitwise_and(img, img, mask = mask)  # save images #cv2.imwrite(name + "_roadarea.png", np.hstack([img, output])) 

there problems code :

firstly, should change

randompixel(roadsample,n) 

to

rpixel=randompixel(roadsample,n) 

even though works because python nice , rpixel delcared beforehand, it's messy : randompixel function returning rpixel , forgetting it.

secondly,the error mention in comment comes double loop. first loop ok, loop every item in rpixel. following line

px = roadsample[i] 

gets corresponding pixel in cropped image, ok. px 3 element array. iterate in need like

for color in px 

not

for b,g,r in px 

because here trying iterate on elements of px @ once. error telling you, "i can't iterate on b value of px, because it's number !" seems you're new programming, don't hesitate use lot of "print" or debug ide see variables @ given time

finally, question, if understood correctly, here proposed code :

import cv2 import numpy np  def bottomcenter(inc,img):     y,x,z = img.shape     h = x/2     x1 = h - (x/inc)     x2 = h + (x/inc)     y1 = y - (y/inc)     y2 = y      bcsample = img[y1:y2, x1:x2]     return(bcsample)  #read image full_road = cv2.imread(fullroad_path)  #crop road part of road there no line inc=10 cropped_road=bottomcenter(inc,road)  #seperate image in each channel (yes, can that, numpy handy !) blue=cropped_road[:,:,0]; green=cropped_road[:,:,1]; red=cropped_road[:,:,2];  #computing mean , std of each channel blue_mean=np.mean(blue); green_mean=np.mean(green); red_mean=np.mean(red); blue_std=np.std(blue); green_std=np.std(green); red_std=np.std(red);  #getting positions of pixels in [color +-std] mask = cv2.inrange(fullroad, np.array([blue_mean-blue_std,green_mean-green_std,red_mean-red_std]), np.array([blue_mean+blue_std,green_mean+green_std,red_mean+red_std])) #masking image output = cv2.bitwise_and(full_road, full_road, mask = mask)