我需要一种算法,该算法可以确定两个图像是否“相似”并识别颜色,亮度,形状等的相似模式。我可能需要一些关于人脑对图像进行“分类”的参数。..
我看过基于hausdorff的匹配,但这似乎主要是为了匹配变形的对象和形状图案。
Answers:
通过使用小波变换将图像分解成签名,我做了类似的事情。
我的方法是从每个转换的通道中选取最重要的n个系数,并记录它们的位置。这是通过根据abs(power)对(power,location)元组的列表进行排序来完成的。相似的图像将具有相似之处,因为它们在相同位置具有显着的系数。
我发现最好将图像转换为YUV格式,这样可以有效地使您在形状(Y通道)和颜色(UV通道)上具有相似的权重。
您可以在mactorii中找到上述的实现,但是不幸的是,我的工作还没有达到我应该做的那么多:-)
我的一些朋友使用的另一种方法取得了令人惊讶的良好效果,它只是简单地将图像调整为4x4像素大小并存储您的签名。通过使用相应的像素计算两个图像之间的曼哈顿距离,可以说出两个图像的相似程度。我没有关于它们如何执行调整大小的详细信息,因此您可能必须尝试使用可用于该任务的各种算法来找到合适的算法。
pHash可能会让您感兴趣。
感知哈希 音频,视频或图像文件的指纹,该指纹在数学上基于其中包含的音频或视觉内容。与依赖于输入中的小变化导致输出中的急剧变化的雪崩效应的加密散列函数不同,如果输入在视觉上或听觉上相似,则感知哈希彼此“接近”。
我已经使用SIFT在不同的图像中重新检测到同一物体。它确实功能强大,但相当复杂,可能会过大。如果假定图像非常相似,则基于两个图像之间的差异的一些简单参数可以告诉您很多信息。一些指针:
我的实验室也需要解决此问题,因此我们使用了Tensorflow。这是用于可视化图像相似性的完整应用程序实现。
有关将图像矢量化以进行相似度计算的教程,请查看此页面。这是Python(同样,请参见该帖子以获取完整的工作流程):
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()
您可以使用感知图像差异
这是一个命令行实用程序,它使用感知指标比较两个图像。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,因此忽略了像素的微小变化。此外,它还大大减少了由随机数生成,操作系统或机器体系结构差异引起的误报数量。
一些图像识别软件解决方案实际上并非纯粹基于算法,而是利用神经网络概念。请访问http://en.wikipedia.org/wiki/Artificial_neural_network,即NeuronDotNet,它也包含有趣的示例:http ://neurondotnet.freehostia.com/index.html
有使用Kohonen神经网络/自组织图的相关研究
出现了更多的学术系统(Google for PicSOM)或较少的学术系统
(http://www.generation5.org/content/2004/aiSomPic.asp(可能不适用于所有工作环境))演示文稿。
这听起来像是视觉问题。您可能需要研究自适应增强以及Burns Line Extraction算法。这两个概念应有助于解决此问题。如果您不熟悉视觉算法,那么边缘检测甚至是一个更简单的起点,因为它说明了基础知识。
至于分类参数:
根据所需的精确度,您可以简单地将图像分解为nxn像素块并进行分析。如果在第一块中获得不同的结果,您将无法停止处理,从而可以提高性能。
为了分析正方形,您可以例如获取颜色值的总和。
我发现本文对解释其工作原理非常有帮助:
http://www.hackerfactor.com/blog/index.php?/archives/432-Looks-Like-It.html
很抱歉在讨论后期加入。
我们甚至可以使用ORB方法来检测两个图像之间的相似特征点。以下链接给出了Python中ORB的直接实现
http://scikit-image.org/docs/dev/auto_examples/plot_orb.html
甚至openCV也可以直接实现ORB。如果您需要更多信息,请遵循下面给出的研究文章。
在另一个线程上有一些很好的答案,但是我想知道涉及频谱分析的东西是否可行?即,将图像分解为相位和幅度信息,然后进行比较。这样可以避免某些与裁剪,转换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。如果您搜索了http://scholar.google.com,我相信您可以针对此提出一些论文。