如何进一步平滑拼图块边缘？Python图像阈值检测优化问询

以下是我用于对图像进行阈值处理以检测拼图块的代码，我希望能够进一步平滑拼图块的边缘。

我从未使用过Python，但为了解决拼图问题，我开展了这个项目，因此需要拼图块的边缘尽可能平滑。

现有代码 #

from PIL import Image, ExifTags, ImageFilter
import numpy as np
import cv2

EXPECTED_PHOTO_ORIENTATION = 1 # Horizontal (normal)

def get_photo_orientation(img):
    exif = img._getexif()
    if exif:
        for tag, value in exif.items():
            if tag in ExifTags.TAGS:
                if ExifTags.TAGS[tag] == 'Orientation':
                    return value
    return None


def binary_pixel_data_for_photo(path, threshold, max_width=None, crop=None):
    """
    Given a bitmap image path, returns a 2D array of 1s and 0s
    crop is either None or (top, right, bottom, left) in pixels
    """
    with Image.open(path) as img:
        if (orientation := get_photo_orientation(img)) is not None and orientation != EXPECTED_PHOTO_ORIENTATION:
            raise Exception(f"Image {path} is not oriented correctly: {orientation}")

        w, h = img.size

        if max_width is not None and img.size[0] > max_width:
            scale_factor = max_width / img.size[0]
            try:
                img = img.resize((max_width, int(img.size[1] * scale_factor)), resample=Image.NEAREST)
            except Exception as e:
                print(f"Error resizing {path}")
                raise e
        else:
            scale_factor = 1.0

        if crop:
            w, h = img.size
            img = img.crop((crop[3], crop[0], w - crop[1], h - crop[2]))


        data, out_w, out_h = threshold_pixels(img, threshold)
        return data, out_w, out_h, scale_factor


def threshold_pixels(img, threshold):
    # Convert image to grayscale numpy array
    grayscale = img.convert('L')
    data = np.array(grayscale)

    # Aplicar convolución
    filtered = cv2.GaussianBlur(data, (9, 9), sigmaX=0)

    # Apply threshold to get binary representation
    binary_data = np.where(filtered <= threshold, 0, 1).astype(np.int8)
    return binary_data, binary_data.shape[1], binary_data.shape[0]


binary_data, out_w, out_h, scale_factor = binary_pixel_data_for_photo("ejemplo.jpeg", 100)

binary_uint8 = binary_data.astype(np.uint8) * 255

contours, _ = cv2.findContours(binary_uint8, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

if contours:
    largest = max(contours, key=cv2.contourArea)
    contour_img = np.zeros_like(binary_uint8)
    cv2.drawContours(contour_img, [largest], 0, 255, -1)

# Mostrar resultado
plt.imshow(cv2.imread("ejemplo.jpeg"))
plt.title("Imagen original")
plt.axis("off")
plt.show()

plt.figure(figsize=(6, 8))
plt.imshow(contour_img)
plt.title("Contorno de la pieza")
plt.axis("off")
plt.show()

处理结果图 #

• 原始图像
  

• 叠加轮廓图
  

• 提取的填充轮廓图
  

内容的提问来源于stack exchange，提问作者Bruno Munné