Revision history [back]

I'm creating an image processor. The intent is to compute the percent of pixels are in range for each color in my swatch. I intend to use the percent to sort my images. For example, if I want blue images, then sort by the blue percent.

My problem is that I'm not getting good percent computations. Images that have very little blue, end up having 17% blue, for example. Can any one help?

Here is my Color class:

class Colour:
  def __init__(self):
    self.swatch = {
        'blue': ((100,168,168), (124,255,255)), # hue, saturation, value
        'cyan': ((88,168,168), (97,255,255)),
        'green': ((50,168,168), (70,255,255)),
        'red': ((88,168,168), (97,255,255)), # invert cyan
        'orange': ((17,168,168), (24,255,255)),
        'yellow': ((25,168,168), (35,255,255)),
        'purple': ((129,168,168), (138,255,255)),
        'pink': ((88,38,168), (97,168,255)), #invert cyan
        'brown': ((17,168,50), (24,255,168)),
        'white': ((0,0,205),(180,50,255)),
        'black': ((0,0,0),(180,255,50))
    }

  def process(self, image, data):
    hsvimg = cv2.cvtColor(image,cv2.COLOR_BGR2HSV)

    for k in self.swatch.keys():
        if k == "red" or k == "pink": #if you are red, invert image and use cyan threshold
            hsvimg = (255-hsvimg)
        inrange = cv2.inRange(hsvimg, self.swatch[k][0], self.swatch[k][1])
        numinrange = cv2.countNonZero(inrange)
        size = hsvimg.shape[0] * hsvimg.shape[1]

        data[k] = float(numinrange) / float(size)

    return data