python 图片相似度计算

最新推荐文章于 2024-09-12 19:51:15 发布

goodstudy@419

最新推荐文章于 2024-09-12 19:51:15 发布

阅读量899

点赞数

分类专栏： python

本文链接：https://blog.csdn.net/qq_41007998/article/details/110393200

版权

python 专栏收录该内容

46 篇文章 0 订阅

订阅专栏

import os
os.environ['KERAS_BACKEND']='tensorflow'
import numpy as np
from keras.applications.vgg16 import VGG16
from keras.applications.vgg16 import preprocess_input
from keras.preprocessing import image
from keras.utils.data_utils import get_file
import ssl
ssl._create_default_https_context = ssl._create_unverified_context # 下载模型的时候不想进行ssl证书校验

# !export TF_CPP_MIN_LOG_LEVEL=2  # 屏蔽警告信息 （这是在jupyter中执行的shell命令，可以去掉这行不影响结果）

# 初始化VGG16模型 include_top=False 是指不保留顶层的3个全连接网络层
model = VGG16(weights='imagenet', include_top=False)

# # 创建模型
#     # Create model.
#     model = Model(inputs, x, name='vgg16')
#
#     # 加载权重
#     # load weights
#     if weights == 'imagenet':
#         if include_top:
#             weights_path = get_file('vgg16_weights_tf_dim_ordering_tf_kernels.h5',
#                                     WEIGHTS_PATH,
#                                     cache_subdir='models')
#         else:
#             weights_path = get_file('vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5',
#                                     WEIGHTS_PATH_NO_TOP,
#                                     cache_subdir='models')
#         model.load_weights(weights_path)
WEIGHTS_PATH = 'https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels.h5'
def get_feature(path):
    """
    VGG16模型抽取图片特征
    """
    # 图片转成PIL的Image对象，并且对图片做了缩放
    img = image.load_img(path, target_size=(224, 224))

    # 图片转成矩阵、并扩充了维度、最后是预处理
    predict_img = preprocess_input(np.expand_dims(image.img_to_array(img),0))

    # 丢入vgg16网络做特征抽取，最后返回特征并展平成一维向量方便计算余弦相似度,flatten是numpy.ndarray.flatten的一个函数，即返回一个一维数组。
    return model.predict(predict_img).flatten()

def cos_sim(a, b):
    """
    计算两个向量之间的余弦相似度
    """
    a = np.mat(a)
    b = np.mat(b)
    return float(a * b.T) / (np.linalg.norm(a) * np.linalg.norm(b))
if __name__ == '__main__':

    image_path = ['D:\\9.jpg', 'D:\\10.jpg']
    ft1, ft2 = [get_feature(p) for p in image_path]

    print(cos_sim(ft1, ft2))