当前位置：首页 > news >正文

Python中提取图片特征的指南

news 来源：原创 2025/6/15 7:44:02

引言

在计算机视觉和图像处理领域，提取图片特征是一项基础而重要的任务。特征提取可以帮助我们理解图像内容，为后续的分类、识别或分析任务提供基础。本文将介绍几种在Python中常用的图片特征提取方法。

准备工作

在开始之前，确保你已经安装了必要的Python库：

pip install opencv-python numpy scikit-learn scikit-image tensorflow pillow

1. 使用OpenCV进行基本特征提取

OpenCV是一个强大的计算机视觉库，提供了多种特征提取方法。

1.1 颜色特征

import cv2
import numpy as npdef extract_color_features(image_path):# 读取图像image = cv2.imread(image_path)image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)# 计算颜色直方图hist = cv2.calcHist([image], [0, 1, 2], None, [8, 8, 8], [0, 256, 0, 256, 0, 256])hist = cv2.normalize(hist, hist).flatten()return hist# 使用示例
features = extract_color_features('example.jpg')
print(f"提取的颜色特征维度: {len(features)}")

1.2 边缘和形状特征

def extract_edge_features(image_path):# 读取图像为灰度图image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)# 使用Canny边缘检测edges = cv2.Canny(image, 100, 200)# 计算边缘直方图hist = cv2.calcHist([edges], [0], None, [256], [0, 256])hist = cv2.normalize(hist, hist).flatten()return hist# 使用示例
edge_features = extract_edge_features('example.jpg')
print(f"提取的边缘特征维度: {len(edge_features)}")

2. 使用scikit-image进行纹理特征提取

scikit-image提供了多种纹理特征提取方法，如局部二值模式(LBP)和灰度共生矩阵(GLCM)。

2.1 局部二值模式(LBP)

from skimage.feature import local_binary_pattern
from skimage import io, colordef extract_lbp_features(image_path):# 读取图像image = io.imread(image_path)gray = color.rgb2gray(image)# 计算LBP特征radius = 3n_points = 8 * radiuslbp = local_binary_pattern(gray, n_points, radius, method='uniform')# 计算LBP直方图hist, _ = np.histogram(lbp, bins=np.arange(0, n_points + 3), range=(0, n_points + 2))hist = hist.astype("float")hist /= (hist.sum() + 1e-6)  # 归一化return hist# 使用示例
lbp_features = extract_lbp_features('example.jpg')
print(f"提取的LBP特征维度: {len(lbp_features)}")

2.2 灰度共生矩阵(GLCM)

from skimage.feature import greycomatrix, greycopropsdef extract_glcm_features(image_path):# 读取图像为灰度图image = io.imread(image_path, as_gray=True)# 计算GLCMglcm = greycomatrix(image, distances=[1], angles=[0], levels=256, symmetric=True, normed=True)# 提取特征contrast = greycoprops(glcm, 'contrast')[0, 0]dissimilarity = greycoprops(glcm, 'dissimilarity')[0, 0]homogeneity = greycoprops(glcm, 'homogeneity')[0, 0]energy = greycoprops(glcm, 'energy')[0, 0]correlation = greycoprops(glcm, 'correlation')[0, 0]return np.array([contrast, dissimilarity, homogeneity, energy, correlation])# 使用示例
glcm_features = extract_glcm_features('example.jpg')
print(f"提取的GLCM特征: {glcm_features}")

3. 使用深度学习提取高级特征

随着深度学习的发展，使用预训练的卷积神经网络(CNN)提取特征变得越来越流行。

3.1 使用预训练的VGG16模型

from tensorflow.keras.applications.vgg16 import VGG16, preprocess_input
from tensorflow.keras.preprocessing import image
from tensorflow.keras.models import Model
import numpy as npdef extract_deep_features(image_path):# 加载预训练的VGG16模型，不包括顶层分类层base_model = VGG16(weights='imagenet')model = Model(inputs=base_model.input, outputs=base_model.get_layer('block5_pool').output)# 加载并预处理图像img = image.load_img(image_path, target_size=(224, 224))x = image.img_to_array(img)x = np.expand_dims(x, axis=0)x = preprocess_input(x)# 提取特征features = model.predict(x)# 展平特征features = features.flatten()return features# 使用示例
deep_features = extract_deep_features('example.jpg')
print(f"提取的深度学习特征维度: {len(deep_features)}")

4. 特征融合

在实际应用中，常常将不同类型的特征融合以获得更好的性能。

def extract_combined_features(image_path):# 提取各种特征color_features = extract_color_features(image_path)edge_features = extract_edge_features(image_path)lbp_features = extract_lbp_features(image_path)glcm_features = extract_glcm_features(image_path)# 尝试提取深度学习特征（可能耗时）try:deep_features = extract_deep_features(image_path)combined_features = np.concatenate([color_features, edge_features, lbp_features, glcm_features, deep_features])except:combined_features = np.concatenate([color_features, edge_features, lbp_features, glcm_features])return combined_features# 使用示例
combined_features = extract_combined_features('example.jpg')
print(f"融合后的特征维度: {len(combined_features)}")