컴퓨터 비전 알고리즘 구현 - 8. 모폴로지 연산

모폴로지 연산

- 컴퓨터 비전의 영역 연산 중 한 종류

- 이미지를 깍거나 팽창시켜준다. -> 침식, 팽창 연산

- 침식 연산과 팽창 연산을 이용하여 열림 연산과 닫힘 연산이 가능.

- 열림 연산 : 침식 -> 팽창. 이미지를 깍아 노이즈를 제거, 깍여진 이미지를 팽창하여 원래 형태로 복원

- 닫힘 연산 : 팽창 ->침식. 이미지를 팽창시켜, 내부 구멍들을 채움, 팽창된 이미지를 침식하여 원래 형태로

 

 

 

 

1. 이미지 로드. 기본 이미지 시각화

 

import numpy as np
import matplotlib.pyplot as plt
from utils import Histogram, Threshold, padding
from PIL import Image


img = Image.open("./res/j.png").convert("L")
img = np.asarray(img)

plt.imshow(img, cmap="gray")

 

 

 

 

 

 

2. 모폴로지 연산 구현

def erosion(boundary=None, kernel=None):
    """
    erosion operation
    - a pixel element is 0 at least under kernel is 0
    -> return 0
    """
    boundary = boundary * kernel
    if (np.min(boundary) == 0):
        return 0
    else:
        return 255

def dilation(boundary=None, kernel=None):
    """
    erosion operation
    - a pixel element is not 0 at least under kernel is not 0
    -> return 255
    """
    boundary = boundary * kernel
    if (np.max(boundary) != 0):
        return 255
    else:
        return 0

def openning(img=None, k_size=None):
    """
    openning operation
    - erosion followed by dilation
    - it removes noise
    """
    erosion_img = morphology(img=img, method=1, k_size=k_size)
    opened_img = morphology(img=erosion_img, method=2, k_size=k_size)
    return opened_img

def closing(img=None, k_size=None):
    """
    closing operation
    - dilation follwed by erosion
    - it can close small holes inside the objects. 
    """
    dilation_img = morphology(img=img, method=2, k_size=k_size)
    closed_img = morphology(img=dilation_img, method=1, k_size=k_size)
    return closed_img


def morphology(img=None, method=None, k_size=None):
    """
    input
    img : input image
    method : 1(erosion), 2(dilation), 3(openning), 4(closing)
    k_size : kernel size
    
    output
    res_img : morphology operation image
    """
    rows, cols = img.shape
    pad_img = padding(img, k_size)
    kernel = np.ones((k_size,k_size))
    res_img = img.copy()
    
    if method == 1 or method == 2:
        for i in range(0, rows):
            for j in range(0, cols):
                if method == 1: #erosion operation
                    res_img[i, j] = erosion(pad_img[i:i+k_size, j:j+k_size], kernel)
                elif method == 2: #
                    res_img[i, j] = dilation(pad_img[i:i+k_size, j:j+k_size], kernel)
    if method == 3:
        res_img = openning(img, k_size=k_size)
    elif method == 4:
        res_img = closing(img, k_size=k_size)

    return res_img

 

 

 

3. 침식 연산 결과

res = morphology(img, method=1, k_size=5)
plt.title("erosion operation with k_size = 5")
plt.imshow(res, cmap="gray")

 

 

 

4. 팽창 연산 결과

res = morphology(img, method=2, k_size=5)
plt.title("dilation operation with k_size = 5")
plt.imshow(res, cmap="gray")

 

 

5. 열기 연산

5.1 열기 연산 이전. 노이즈 이미지

 

noised_j = Image.open("./res/noised_j.PNG").convert("L")
noised_j = np.asarray(noised_j)
plt.title("befor opening operation")
plt.imshow(noised_j, cmap="gray")

 

5.2 열기 연산 수행

res = morphology(noised_j, method=3, k_size=7)
plt.title("after opening operation with k_size :7")
plt.imshow(res, cmap="gray")

 

 

 

6. 닫기 연산

6.1 닫기 연산 전. 이미지

 

small_hole = Image.open("./res/j_small_hole.PNG").convert("L")
small_hole = np.asarray(small_hole)
plt.title("before closing operation")
plt.imshow(small_hole, cmap="gray")

 

 

6.2 닫기 연산 후

res = morphology(small_hole, method=4, k_size=5)
plt.title("after closing operation with k_size :7")
plt.imshow(res, cmap="gray")