寻找相似图像的算法

我也做过类似的事情,通过使用 wavelet transform .

我的方法是选择最重要的 来自每个变换通道的系数,并记录其位置。这是通过根据abs(power)对(power,location)元组列表进行排序来完成的。相似的图像将共享相似性,因为它们在相同的位置具有显著的系数。

您可以在中找到我对上述内容的实现 mactorii ,不幸的是,我没有像我应该做的那样努力:-)

我的一些朋友使用的另一种方法效果出人意料地好,就是简单地将图像大小调整为4x4像素,并存储为您的签名。两幅图像的相似程度可以通过计算 Manhattan distance 在两幅图像之间,使用相应的像素。我没有他们如何执行调整大小的详细信息,因此您可能需要使用该任务可用的各种算法来找到合适的算法。

pHash 你可能会感兴趣。

SIFT 重新检测不同图像中的同一对象。它确实很强大,但相当复杂,可能有点过头了。如果图像应该非常相似,那么基于两个图像之间差异的一些简单参数可以告诉您很多信息。一些建议:

• 每个通道标准化图像上的色差总和。
• 查找图像中的边缘并测量两幅图像中边缘像素之间的距离。(用于形状)
• 将图像分成一组离散区域,并比较每个区域的平均颜色。
• 对一个(或一组)级别的图像设置阈值,并计算结果黑白图像不同的像素数。

我的实验室也需要解决这个问题,我们使用了Tensorflow。这里有一个 full app 可视化图像相似性的实现。

有关将图像矢量化以进行相似性计算的教程,请查看 this page

from __future__ import absolute_import, division, print_function

"""

This is a modification of the classify_images.py
script in Tensorflow. The original script produces
string labels for input images (e.g. you input a picture
of a cat and the script returns the string "cat"); this
modification reads in a directory of images and 
generates a vector representation of the image using
the penultimate layer of neural network weights.

Usage: python classify_images.py "../image_dir/*.jpg"

"""

# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================

"""Simple image classification with Inception.

Run image classification with Inception trained on ImageNet 2012 Challenge data
set.

This program creates a graph from a saved GraphDef protocol buffer,
and runs inference on an input JPEG image. It outputs human readable
strings of the top 5 predictions along with their probabilities.

Change the --image_file argument to any jpg image to compute a
classification of that image.

Please see the tutorial and website for a detailed description of how
to use this script to perform image recognition.

https://tensorflow.org/tutorials/image_recognition/
"""

import os.path
import re
import sys
import tarfile
import glob
import json
import psutil
from collections import defaultdict
import numpy as np
from six.moves import urllib
import tensorflow as tf

FLAGS = tf.app.flags.FLAGS

# classify_image_graph_def.pb:
#   Binary representation of the GraphDef protocol buffer.
# imagenet_synset_to_human_label_map.txt:
#   Map from synset ID to a human readable string.
# imagenet_2012_challenge_label_map_proto.pbtxt:
#   Text representation of a protocol buffer mapping a label to synset ID.
tf.app.flags.DEFINE_string(
    'model_dir', '/tmp/imagenet',
    """Path to classify_image_graph_def.pb, """
    """imagenet_synset_to_human_label_map.txt, and """
    """imagenet_2012_challenge_label_map_proto.pbtxt.""")
tf.app.flags.DEFINE_string('image_file', '',
                           """Absolute path to image file.""")
tf.app.flags.DEFINE_integer('num_top_predictions', 5,
                            """Display this many predictions.""")

# pylint: disable=line-too-long
DATA_URL = 'http://download.tensorflow.org/models/image/imagenet/inception-2015-12-05.tgz'
# pylint: enable=line-too-long


class NodeLookup(object):
  """Converts integer node ID's to human readable labels."""

  def __init__(self,
               label_lookup_path=None,
               uid_lookup_path=None):
    if not label_lookup_path:
      label_lookup_path = os.path.join(
          FLAGS.model_dir, 'imagenet_2012_challenge_label_map_proto.pbtxt')
    if not uid_lookup_path:
      uid_lookup_path = os.path.join(
          FLAGS.model_dir, 'imagenet_synset_to_human_label_map.txt')
    self.node_lookup = self.load(label_lookup_path, uid_lookup_path)

  def load(self, label_lookup_path, uid_lookup_path):
    """Loads a human readable English name for each softmax node.

    Args:
      label_lookup_path: string UID to integer node ID.
      uid_lookup_path: string UID to human-readable string.

    Returns:
      dict from integer node ID to human-readable string.
    """
    if not tf.gfile.Exists(uid_lookup_path):
      tf.logging.fatal('File does not exist %s', uid_lookup_path)
    if not tf.gfile.Exists(label_lookup_path):
      tf.logging.fatal('File does not exist %s', label_lookup_path)

    # Loads mapping from string UID to human-readable string
    proto_as_ascii_lines = tf.gfile.GFile(uid_lookup_path).readlines()
    uid_to_human = {}
    p = re.compile(r'[n\d]*[ \S,]*')
    for line in proto_as_ascii_lines:
      parsed_items = p.findall(line)
      uid = parsed_items[0]
      human_string = parsed_items[2]
      uid_to_human[uid] = human_string

    # Loads mapping from string UID to integer node ID.
    node_id_to_uid = {}
    proto_as_ascii = tf.gfile.GFile(label_lookup_path).readlines()
    for line in proto_as_ascii:
      if line.startswith('  target_class:'):
        target_class = int(line.split(': ')[1])
      if line.startswith('  target_class_string:'):
        target_class_string = line.split(': ')[1]
        node_id_to_uid[target_class] = target_class_string[1:-2]

    # Loads the final mapping of integer node ID to human-readable string
    node_id_to_name = {}
    for key, val in node_id_to_uid.items():
      if val not in uid_to_human:
        tf.logging.fatal('Failed to locate: %s', val)
      name = uid_to_human[val]
      node_id_to_name[key] = name

    return node_id_to_name

  def id_to_string(self, node_id):
    if node_id not in self.node_lookup:
      return ''
    return self.node_lookup[node_id]


def create_graph():
  """Creates a graph from saved GraphDef file and returns a saver."""
  # Creates graph from saved graph_def.pb.
  with tf.gfile.FastGFile(os.path.join(
      FLAGS.model_dir, 'classify_image_graph_def.pb'), 'rb') as f:
    graph_def = tf.GraphDef()
    graph_def.ParseFromString(f.read())
    _ = tf.import_graph_def(graph_def, name='')


def run_inference_on_images(image_list, output_dir):
  """Runs inference on an image list.

  Args:
    image_list: a list of images.
    output_dir: the directory in which image vectors will be saved

  Returns:
    image_to_labels: a dictionary with image file keys and predicted
      text label values
  """
  image_to_labels = defaultdict(list)

  create_graph()

  with tf.Session() as sess:
    # Some useful tensors:
    # 'softmax:0': A tensor containing the normalized prediction across
    #   1000 labels.
    # 'pool_3:0': A tensor containing the next-to-last layer containing 2048
    #   float description of the image.
    # 'DecodeJpeg/contents:0': A tensor containing a string providing JPEG
    #   encoding of the image.
    # Runs the softmax tensor by feeding the image_data as input to the graph.
    softmax_tensor = sess.graph.get_tensor_by_name('softmax:0')

    for image_index, image in enumerate(image_list):
      try:
        print("parsing", image_index, image, "\n")
        if not tf.gfile.Exists(image):
          tf.logging.fatal('File does not exist %s', image)

        with tf.gfile.FastGFile(image, 'rb') as f:
          image_data =  f.read()

          predictions = sess.run(softmax_tensor,
                          {'DecodeJpeg/contents:0': image_data})

          predictions = np.squeeze(predictions)

          ###
          # Get penultimate layer weights
          ###

          feature_tensor = sess.graph.get_tensor_by_name('pool_3:0')
          feature_set = sess.run(feature_tensor,
                          {'DecodeJpeg/contents:0': image_data})
          feature_vector = np.squeeze(feature_set)        
          outfile_name = os.path.basename(image) + ".npz"
          out_path = os.path.join(output_dir, outfile_name)
          np.savetxt(out_path, feature_vector, delimiter=',')

          # Creates node ID --> English string lookup.
          node_lookup = NodeLookup()

          top_k = predictions.argsort()[-FLAGS.num_top_predictions:][::-1]
          for node_id in top_k:
            human_string = node_lookup.id_to_string(node_id)
            score = predictions[node_id]
            print("results for", image)
            print('%s (score = %.5f)' % (human_string, score))
            print("\n")

            image_to_labels[image].append(
              {
                "labels": human_string,
                "score": str(score)
              }
            )

        # close the open file handlers
        proc = psutil.Process()
        open_files = proc.open_files()

        for open_file in open_files:
          file_handler = getattr(open_file, "fd")
          os.close(file_handler)
      except:
        print('could not process image index',image_index,'image', image)

  return image_to_labels


def maybe_download_and_extract():
  """Download and extract model tar file."""
  dest_directory = FLAGS.model_dir
  if not os.path.exists(dest_directory):
    os.makedirs(dest_directory)
  filename = DATA_URL.split('/')[-1]
  filepath = os.path.join(dest_directory, filename)
  if not os.path.exists(filepath):
    def _progress(count, block_size, total_size):
      sys.stdout.write('\r>> Downloading %s %.1f%%' % (
          filename, float(count * block_size) / float(total_size) * 100.0))
      sys.stdout.flush()
    filepath, _ = urllib.request.urlretrieve(DATA_URL, filepath, _progress)
    print()
    statinfo = os.stat(filepath)
    print('Succesfully downloaded', filename, statinfo.st_size, 'bytes.')
  tarfile.open(filepath, 'r:gz').extractall(dest_directory)


def main(_):
  maybe_download_and_extract()
  if len(sys.argv) < 2:
    print("please provide a glob path to one or more images, e.g.")
    print("python classify_image_modified.py '../cats/*.jpg'")
    sys.exit()

  else:
    output_dir = "image_vectors"
    if not os.path.exists(output_dir):
      os.makedirs(output_dir)

    images = glob.glob(sys.argv[1])
    image_to_labels = run_inference_on_images(images, output_dir)

    with open("image_to_labels.json", "w") as img_to_labels_out:
      json.dump(image_to_labels, img_to_labels_out)

    print("all done")
if __name__ == '__main__':
  tf.app.run()

你可以用 Perceptual Image Diff

它是一个命令行实用程序,使用感知度量比较两个图像。也就是说,它使用人类视觉系统的计算模型来确定两个图像在视觉上是否不同,因此忽略像素的微小变化。此外,它还大大减少了由于随机数生成、操作系统或机器体系结构差异而导致的误报次数。

希望这有帮助!

一些图像识别软件解决方案实际上并不是纯粹基于算法,而是利用 神经网络 取而代之的是概念。退房 http://en.wikipedia.org/wiki/Artificial_neural_network 即NeuronDotNet,其中还包括有趣的样本: http://neurondotnet.freehostia.com/index.html

使用Kohonen神经网络/自组织映射进行相关研究

更多学术系统(谷歌用于PicSOM)或更少学术系统
( http://www.generation5.org/content/2004/aiSomPic.asp ,(可能不合适 对于所有工作环境,都有演示文稿。

计算大幅缩小版本(例如:6x6像素)像素颜色值差异的平方和效果很好。相同的图像产生0,相似的图像产生小数字,不同的图像产生大数字。

上面的其他人首先提出的闯入YUV的想法听起来很有趣——虽然我的想法很有效,但我希望我的图像被计算为“不同的”,以便产生正确的结果——即使是从色盲观察者的角度。

关于分类的参数:

• 调色板及;位置(梯度计算、颜色直方图)
• 包含的形状(Ada.检测形状的增强/培训)

根据需要的精确结果,您可以简单地将图像分为n x n像素块并进行分析。如果在第一个块中得到不同的结果,则无法停止处理,从而导致一些性能改进。

例如,为了分析正方形,您可以获得颜色值的总和。

我发现这篇文章非常有助于解释它的工作原理:

http://www.hackerfactor.com/blog/index.php?/archives/432-Looks-Like-It.html

您可以在两幅图像之间执行某种块匹配运动估计,并测量残差和运动矢量成本的总和(就像在视频编码器中一样)。这将补偿运动;对于奖励点,执行仿射变换运动估计(补偿缩放和拉伸等)。你也可以做重叠块或光流。

作为第一步,您可以尝试使用颜色直方图。然而,你确实需要缩小你的问题范围。通用图像匹配是一个非常困难的问题。

为参加讨论的时间较晚表示歉意。

我们甚至可以使用ORB方法来检测两幅图像之间的相似特征点。

http://scikit-image.org/docs/dev/auto_examples/plot_orb.html

甚至openCV也直接实现了ORB。如果你想了解更多信息,请阅读下面的研究文章。

https://www.researchgate.net/publication/292157133_Image_Matching_Using_SIFT_SURF_BRIEF_and_ORB_Performance_Comparison_for_Distorted_Images

关于这一点,在另一条线索中有一些很好的答案,但我想知道涉及光谱分析的东西是否可行?即,将图像分解为其相位和振幅信息,并进行比较。这可以避免一些与种植、转化和强度差异有关的问题。无论如何,这只是我的猜测,因为这似乎是一个有趣的问题。如果你搜索 http://scholar.google.com 我相信你能拿出几份关于这个的文件。