I have developed a convolution with a kernel set that reliably removes singular and protruding white dots when applied 2-3 times to an image of zeros (black) and ones (white):

import numpy as np
from PIL import Image
from scipy.ndimage import convolve

kernels = [
    np.array([[1, 1, 1], [-4, 1, -4], [-4, -4, -4]]),
    np.array([[-4, -4, 1], [-4, 1, 1], [-4, -4, 1]]),
    np.array([[-4, -4, -4], [-4, 1, -4], [1, 1, 1]]),
    np.array([[1, -4, -4], [1, 1, -4], [1, -4, -4]])
]

image_path = 'path/to/image/file'
with Image.open(image_path) as image:
    
    if image.mode == 'RGBA':
        image = image.convert('RGB')

    image_np = np.array(image)
    mask = np.isin(image_np, [200, 200, 200]).all(axis=-1)
    
    binary_mask = np.zeros(image_np.shape[:2], dtype=int)
    binary_mask[mask] = 1
    
    for kernel in kernels:
        convolution = convolve(binary_mask, kernel, mode='constant')
        binary_mask[(binary_mask == 1) & (convolution >= 1)] = 0

The most important property of the convolution is that it preserves closed lines (e.g. the edge of a rectangle). However, the convolution only removes open lines (i.e. lines with two ends) completely if the convolution is performed half as often as the length of the line.

This is very ineffective for long lines. Can you help me to develop a kernel (set) that removes open 1-pixel-thick lines after a few convolutions, but preserves closed 1-pixel-thick lines?

解决方案

暂无回答