dog_extrema.py

import matplotlib.pyplot as plt 
import cv2
import numpy as np

def GetNeighbors(r, c, A, B, C):
    # 1 2 3 
    # 4 5 6 
    # 7 8 9 
    n = [(-1, -1), (-1, 0), (-1, 1), (0, -1), (0, 0), (0, 1), (1, -1), (1, 0), (1, 1)]
    neighbors = []
    for p in n:
        neighbors.append(A[r + p[0], c + p[1]])
        neighbors.append(C[r + p[0], c + p[1]])
        if not (p[0] == 0 and p[1] == 0):
            neighbors.append(B[r + p[0], c + p[1]])
    return neighbors

def IsExtrema(me, neighbors):
    return (me > neighbors).all() or (me < neighbors).all()    

def FindExtremas(A, B, C):
    rows, cols = B.shape[:2]
    ret = []
    for i in xrange(1, rows-1):
        for j in xrange(1, cols-1):
            val = B[i, j]
            neighbors = GetNeighbors(i, j, A, B, C)
            if IsExtrema(val, neighbors):
                ret.append((i, j))
    return ret 

def DoG(I, bigK, smallK):
    blurBig = cv2.GaussianBlur(img,(bigK,bigK),0)
    blurSamll = cv2.GaussianBlur(img,(smallK,smallK),0)
    return blurBig - blurSamll
    
def main():
    I = np.ones((400, 400), dtype = np.float32) * 255
    
    I[20:80, 20:80] = 0 # draw a black rectangle
    
    I[200:280, 200:280] = 0 # draw a black rectangle

    cv2.circle(I, (130, 130), 40, (0,0,0), -1)
    
    # create blured image with kernel: 3,5,7,9
    L3 = cv2.GaussianBlur(I,(3,3),0)
    L5 = cv2.GaussianBlur(I,(5,5),0)
    L7 = cv2.GaussianBlur(I,(7,7),0)
    L9 = cv2.GaussianBlur(I,(9,9),0)
    
    # construct DoG 

    L35 = L3 - L5 
    L57 = L5 - L7 
    L79 = L7 - L9
    
    extremas = FindExtremas(L35, L57, L79)
    bgr = cv2.cvtColor(I, cv2.COLOR_GRAY2BGR)
    
    for p in extremas:
        cv2.circle(bgr, (p[1], p[0]), 2, (0,0,255), 2)
                
    cv2.imshow("bgr", bgr)
    cv2.waitKey(0)
    
    
if __name__ == '__main__':
    main()