基于Python做形状相似性判断的实现方法

import cv2
import numpy as np

# 读取图像并二值化
image = cv2.imread('shape.png', cv2.IMREAD_GRAYSCALE)
_, binary = cv2.threshold(image, 127, 255, cv2.THRESH_BINARY)

# 提取轮廓
contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

# 计算Hu矩
moments = cv2.moments(contours[0])
hu_moments = cv2.HuMoments(moments).flatten()

# 相似度计算（欧氏距离）
def compare_hu(hu1, hu2):
    return np.linalg.norm(hu1 - hu2)

# 示例比较
hu2 = cv2.HuMoments(cv2.moments(contours[1])).flatten()
similarity = compare_hu(hu_moments, hu2)
print(f"Hu矩相似度：{1/(1+similarity):.2f}")  # 值越大越相似

from skimage.measure import find_contours
import numpy as np

# 提取轮廓点
contour = find_contours(binary, level=0.8)[0]

# 构建形状上下文描述符
def shape_context(contour, n_bins=5):
    n_points = len(contour)
    dist_matrix = np.sqrt(np.sum((contour[:, np.newaxis] - contour)**2, axis=-1))
    angle_matrix = np.arctan2(contour[:, 0][:, np.newaxis] - contour[:, 0], 
                             contour[:, 1][:, np.newaxis] - contour[:, 1])
    
    # 构建直方图
    hist = np.zeros((n_points, n_bins, n_bins))
    for i in range(n_points):
        dist_bins = np.linspace(0, dist_matrix[i].max(), n_bins+1)
        angle_bins = np.linspace(-np.pi, np.pi, n_bins+1)
        hist[i] = np.histogram2d(dist_matrix[i], angle_matrix[i], 
                                bins=[dist_bins, angle_bins])[0]
    return hist.flatten()

# 相似度计算（余弦相似度）
from scipy.spatial.distance import cosine
desc1 = shape_context(contour)
desc2 = shape_context(another_contour)
similarity = 1 - cosine(desc1, desc2)

import cv2

# 创建Hausdorff距离提取器
matcher = cv2.createHausdorffDistanceExtractor()

# 计算轮廓间距离
d1 = matcher.computeDistance(contours[0], contours[1])
d2 = matcher.computeDistance(contours[1], contours[0])
hausdorff = max(d1, d2)

print(f"Hausdorff距离：{hausdorff:.2f}（值越小越相似）")

import cv2

# 创建SIFT检测器
sift = cv2.SIFT_create()

# 检测关键点与描述符
kp1, desc1 = sift.detectAndCompute(image, None)
kp2, desc2 = sift.detectAndCompute(another_image, None)

# 特征匹配
bf = cv2.BFMatcher(cv2.NORM_L2)
matches = bf.match(desc1, desc2)

# 按距离排序
matches = sorted(matches, key=lambda x: x.distance)
similarity = len(matches)  # 匹配点数量作为相似度

from sklearn.svm import SVC
from sklearn.metrics import accuracy_score

# 提取多个样本的Hu矩
hu_features = []
for sample in samples:
    moments = cv2.moments(cv2.findContours(...)[0])
    hu = cv2.HuMoments(moments).flatten()
    hu_features.append(hu)

# 训练分类器
clf = SVC(kernel='linear')
clf.fit(hu_features, labels)

# 预测与评估
preds = clf.predict(test_features)
print(f"准确率：{accuracy_score(test_labels, preds):.2f}")