图像相似度搜索：基于SIFT特征和颜色直方图的KNN方法

import cv2
import numpy as np
import os
from sklearn.decomposition import PCA

# 图像集文件夹路径
IMAGE_FOLDER_PATH = './100/'

# 需要检索的图片路径
QUERY_IMAGE_PATH = './01.jpg'

# KNN参数
KNN_K = 3

# 颜色直方图参数
HISTOGRAM_SIZE = [4, 4, 4]
# HISTOGRAM_SIZE = [8, 8, 8]
# PCA参数
PCA_COMPONENTS = 5
# PCA_COMPONENTS = 300

def read_images(image_folder_path):
    '''
    读取并返回图像集
    :param image_folder_path: 图像集文件夹路径
    :return: 图像集列表
    '''
    images = []
    for file_name in os.listdir(image_folder_path):
        if file_name.endswith('.jpg'):
            file_path = os.path.join(image_folder_path, file_name)
            images.append(cv2.imread(file_path))
    return images


def extract_sift_features(image):
    '''
    提取图像SIFT特征
    :param image: 输入图像
    :return: 特征向量
    '''
    gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    sift = cv2.xfeatures2d.SIFT_create()
    keypoints, descriptors = sift.detectAndCompute(gray_image, None)
    return descriptors


def compute_color_histogram(image, histogram_size):
    '''
    计算图像颜色直方图特征向量
    :param image: 输入图像
    :param histogram_size: 直方图尺寸
    :return: 特征向量
    '''
    histogram = cv2.calcHist([image], [0, 1, 2], None, histogram_size, [0, 256, 0, 256, 0, 256])
    histogram = cv2.normalize(histogram, None).flatten()
    return histogram


def run_pca(features, components):
    '''
    使用PCA方法进行特征降维
    :param features: 待降维特征
    :param components: 保留主成分数量
    :return: 降维后的特征
    '''
    pca = PCA(n_components=components)
    pca.fit(features)
    reduced_features = pca.transform(features)
    return reduced_features


# 读取图像集
image_set = read_images(IMAGE_FOLDER_PATH)

# 提取SIFT和颜色直方图特征
sift_features = []
color_histograms = []
for image in image_set:
    sift_features.append(extract_sift_features(image))
    color_histograms.append(compute_color_histogram(image, HISTOGRAM_SIZE))

# 转换特征向量形状
sift_features = np.vstack(sift_features)
color_histograms = np.vstack(color_histograms)


# # 对SIFT特征进行PCA降维
# sift_features = run_pca(sift_features, PCA_COMPONENTS)

# 对SIFT特征进行PCA降维
sift_features = run_pca(sift_features, PCA_COMPONENTS)
query_sift_features = run_pca(query_sift_features, PCA_COMPONENTS)


# 读取要查询的图像
query_image = cv2.imread(QUERY_IMAGE_PATH)

# 提取SIFT和颜色直方图特征
query_sift_features = extract_sift_features(query_image)
query_color_histogram = compute_color_histogram(query_image, HISTOGRAM_SIZE)

# 调整特征向量形状
query_sift_features = query_sift_features.reshape((1, -1))

# 对SIFT特征进行PCA降维
query_sift_features = run_pca(query_sift_features, PCA_COMPONENTS)

# 计算待查询图像的SIFT和颜色直方图特征与图像集中所有图像的距离
sift_distances = np.linalg.norm(sift_features - query_sift_features, axis=1)
color_histogram_distances = np.linalg.norm(color_histograms - query_color_histogram, axis=1)

# 合并SIFT和颜色直方图特征的距离结果
distances = sift_distances + color_histogram_distances

# 对距离结果进行排序，并获取前K个相似的图像索引
sorted_index = np.argsort(distances)[:KNN_K]

# 显示结果
for i in sorted_index:
    cv2.imshow(f'Result-{i}', image_set[i])
cv2.waitKey(0)
cv2.destroyAllWindows()

n_components=128 must be between 0 and min(n_samples, n_features)=1 with svd_solver='full'
如何解决这个问题内容：这个问题是由于PCA的n_components参数设置错误导致的，可能是设置的值超出了特征向量的维度范围。需要根据实际情况重新设置n_components参数，确保其不超出特征向量的维度范围。如果无法确定合适的n_components值，可以使用默认值或者试着使用其他降维方法。