Random_walker segmentation for color image

Hello,
I am implementing the Random-Walker segmentation for color images in order to segment the leaves in images. I have read the whole code of Random-Walker Segmentation for gray image, and now I have a lot of problem in implementing the code for color images. Do you have any idea to help me? My DSC is really low ( about 0.22) and I want to improve it. Here is my code:
def segleaf(I):

I = util.img_as_float(I) 
I_rescaled = rescale_intensity (I)
green = I_rescaled[:,:,1] / (np.sum(I_rescaled, axis=2) + .0000001)
seeds = np. zeros(green.shape, dtype =np. uint )     
seeds [ green[:,:]> 0.5 ] = 2
seeds [ green[:,:]<=0.25] = 1  

Labels =random_walker ( I , seeds , beta =10, copy=False, multichannel= True)
L= morphology.remove_small_objects(Labels, min_size=2 ) 
L=clear_border(L, buffer_size=40, bgval=0)
return L

my results:

DSC for image_0002.png: 0.22927967607832228
download