vessel_segmentation_part1

import numpy as np
import cv2

####normalizing image
def normalize(img):
    max_val=np.amax(img)
    r,c=img.shape
    for i in xrange(r):
        for j in xrange(c):
            if (img[i][j]==0):
                img[i][j]=0;
            else:
                img[i][j]=img[i][j]/max_val


    return(img)

####getDoG
def getDoG(img,sigma,onoff,sigmaRatio,width,th):
    ##create diff of Gaussian kernal
    sz=(sigma*3)*2+1
    g1 = fspecial((sz,sz),sigma)
    g2 = fspecial((sz,sz),sigma*sigmaRatio);

    if (onoff==1):
        G=g2-g1;  ### difference of guassians
    else:
        G=g1-g2;

    print(np.amax(G))
    
       
    dst = cv2.filter2D(img*255,-1,G) ###convolution of image
    return(dst)

def DoGBankResponse(img,params_sigmalist,params_polaritylist):
    sz1,sz2=img.shape;
    DoGBank=np.zeros((sz1,sz2,params_sigmalist,params_polaritylist),dtype=np.uint8)
    for i in xrange(params_sigmalist):
        for j in xrange(params_polaritylist):
            DoGBank[:,:,i,j]=getDoG(img,3,1,0.5,0,0)  ########change paramenters
            cv2.namedWindow('w', cv2.WINDOW_NORMAL)
            cv2.imshow("w",DoGBank[:,:,i,j]); 
    #print(DoGBank[:,:,0,0].shape)
    #cv2.namedWindow('w', cv2.WINDOW_NORMAL)
    #cv2.imshow("w",DoGBank[:,:,0,0]);  
 #   return(DoGBank);



def fspecial(shape,sigma):
    """
    2D gaussian mask - should give the same result as MATLAB's
    fspecial('gaussian',[shape],[sigma])
    """
    m,n = [(ss-1.)/2. for ss in shape]
    y,x = np.ogrid[-m:m+1,-n:n+1]
    h = np.exp( -(x*x + y*y) / (2.*sigma*sigma) )
    h[ h < np.finfo(h.dtype).eps*h.max() ] = 0
    sumh = h.sum()
    if sumh != 0:
        h /= sumh
    return h    
    

####proto type pattern

x=100;
y=100;
line1=np.zeros((201,201), dtype=np.uint8);
line1[:,x]=255;
line2=normalize(line1);

DoGBankResponse(line2,1,1)

cv2.waitKey(0);