refactor(hsv): refactor hsv image search

Author: zibuyu1995

imageSearchEngineWithPython/.gitignore hsv_image_search/.gitignore

@@ -0,0 +1,76 @@
+# 基于HSV颜色特征的图像搜索引擎
+
+### DP:
+* 编写时间: 2018-11-15
+* 编写人: Mousse
+* 联系邮箱: zibuyu1995@gmail.com
+
+### 系统环境:
+* python3.6
+* opencv 3.4.2
+
+### 文件结构:
+```
+├── README.md
+├── app.py  # flask 运行代码
+├── config
+│   ├── __init__.py
+│   └── config.py  # 配置文件
+├── generate_index.py  # 索引文件生成
+├── mlibs
+│   ├── __init__.py
+│   ├── hsv_features.py # HSV 特征提取
+│   └── image_match.py  # HSV 特征匹配
+├── static
+│   ├── dataset  # 数据集
+├── dataset.db  # 序列化后数据集(需要执行generate_index.py后生成)
+├── templates
+│   ├── index.html
+│   └── search.html
+└── test
+    └── 100900.png  # 测试图片
+└── requirements.txt # 依赖
+```
+
+### 项目运行:
+> 创建虚拟环境后请依据场景修改config/config.py 配置
+
+* 安装依赖
+```bash
+pip install -r requirements.txt -i https://pypi.douban.com/simple/
+```
+
+* 生成索引文件
+```bash
+python generate_index.py
+```
+
+* 运行项目
+```bash
+# 常规运行
+python app.py
+# gunicorn 运行
+gunicorn --workers=2 --bind=0.0.0.0:5555 app:app
+```
+
+### 测试:
+> GET http://0.0.0.0:5555/
+
+* 环境: macOS
+![image](https://user-images.githubusercontent.com/17525759/48668695-1b5efb80-eb2f-11e8-895b-4c9c4c1a6105.png)
+
+### 常见错误处理:
+* may have been in progress in another thread when fork() was called.
+```bash
+export OBJC_DISABLE_INITIALIZE_FORK_SAFETY=YES
+```
+
+### 备注:
+* HSV 算法不适合于颜色不鲜明的图像匹配
+* image_bins: （8, 3, 3）：最佳颜色特征点选择
+* 有什么问题请联系我: zibuyu1995@gmail.com
+
+### reference
+* 数据集下载: https://pan.baidu.com/s/1bnhR65SyfONoKTK90T57tQ 密码:r0d2
+* [flask](http://flask.pocoo.org/docs/0.12/)
+

hsv_image_search/README.md

@@ -0,0 +1,83 @@
+#!/usr/bin/env python
+# coding=utf-8
+
+import time
+from operator import itemgetter
+
+from flask import Flask, render_template, request, abort
+from tinydb import TinyDB
+from werkzeug.contrib.cache import SimpleCache
+
+from config.config import (
+    image_bins, site_name, upload_image_path
+)
+from mlibs.hsv_features import get_image_feature
+from mlibs.image_match import chi_square
+
+
+app = Flask(__name__)
+cache = SimpleCache()
+
+
+@app.route('/')
+def index():
+    return render_template('index.html', site_name=site_name)
+
+
+@app.route("/search")
+def view_search_result():
+    response_dict = cache.get("response_dict")
+    if not response_dict:
+        response_dict = {}
+    return render_template('search.html', **response_dict)
+
+
+@app.route("/search", methods=["POST"])
+def new_search():
+    upload_image = request.files.get("image")
+    if not upload_image:
+        raise abort(404)
+    image_types = ['image/jpeg', 'image/jpg', 'image/png']
+    if upload_image.content_type not in image_types:
+        raise abort(404)
+    upload_image_type = upload_image.content_type.split('/')[-1]
+    file_name = str(time.time())[:10] + '.' + upload_image_type
+    file_path = upload_image_path + file_name
+    upload_image.save(file_path)
+    search_results = image_search(file_path)[:5]
+    response_dict = {
+        'site_name': site_name,
+        'upload_image': file_name,
+        'search_results': search_results
+    }
+    cache.set("response_dict", response_dict)
+    return render_template('search.html', **response_dict)
+
+
+def image_search(image_path=None):
+    """ 图像搜索 """
+    global cache_features
+
+    match_results = []
+    match_results_append = match_results.append
+    search_feature = get_image_feature(image_path, image_bins)[1]
+    for image_uid, feature in cache_features.items():
+        distance = chi_square(feature, search_feature)
+        match_results_append((image_uid, distance))
+    match_results = sorted(match_results, key=itemgetter(1))
+    return match_results
+
+
+if __name__ != '__main__':
+    # gunicorn 运行
+    db = TinyDB('./static/dataset.db')
+    table = db.table('feature')
+    cache_features = table.all()[0]
+
+
+if __name__ == '__main__':
+    db = TinyDB('./static/dataset.db')
+    table = db.table('feature')
+    cache_features = table.all()[0]
+    app.run(host='0.0.0.0', port=5555)
+

hsv_image_search/app.py

@@ -0,0 +1,13 @@
+# coding: utf-8
+
+import os
+from multiprocessing import cpu_count
+
+site_name = '基于HSV图像搜索引擎'  # 网站标题
+project_path = os.path.abspath(os.path.join(os.path.dirname(__file__), os.pardir))
+cpu_count = cpu_count()
+
+upload_image_path = os.path.join(project_path, 'static/uploads/')  # 上传图片保存地址
+dataset_db_path = os.path.join(project_path, 'static/dataset.db')  # 序列化后数据集保存地址
+dataset_path = os.path.join(project_path, 'static/dataset/')  # 数据集图片保存地址
+image_bins = (8, 3, 3)  # HSV所占权重

ORB-imageSearch/config/__init__.py hsv_image_search/config/__init__.py

@@ -0,0 +1,59 @@
+#!/usr/bin/env python
+# coding=utf-8
+
+"""
+Time    : 2018-11-18
+Author  : Mousse
+email: zibuyu1995@gmail.com
+"""
+
+import asyncio
+import glob
+import ujson
+from concurrent.futures import ProcessPoolExecutor
+
+import uvloop
+from tinydb import TinyDB
+
+from config.config import dataset_path, cpu_count, image_bins
+from mlibs.hsv_features import get_image_feature
+
+
+asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
+
+
+def feature_persistence(task_results: list) -> int:
+    """ 图像特征集持久化到key-value数据库 """
+    feature_dict = dict(task_result.result() for task_result in task_results)
+    db = TinyDB('./static/dataset.db')
+    table = db.table('feature')
+    table.insert(feature_dict)
+    feature_count = len(feature_dict)
+    return feature_count
+
+
+async def generate_image_index(eve_loop, processes_executor):
+    """ 多进程生成图像索引 """
+    image_feature_tasks = []
+    task_append = image_feature_tasks.append
+    for image_path in glob.glob(dataset_path + "/*.png"):
+        task_append(
+            eve_loop.run_in_executor(
+                processes_executor, get_image_feature, image_path, image_bins
+            )
+        )
+    task_results, _ = await asyncio.wait(image_feature_tasks)
+    feature_count = feature_persistence(task_results)
+    print(f"{feature_count}幅图像完成索引")
+
+
+if __name__ == '__main__':
+    asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
+    executor = ProcessPoolExecutor(max_workers=cpu_count)
+    event_loop = asyncio.get_event_loop()
+    try:
+        event_loop.run_until_complete(
+            generate_image_index(event_loop, executor)
+        )
+    finally:
+        event_loop.close()

hsv_image_search/config/config.py

@@ -0,0 +1,66 @@
+# coding=utf-8
+
+
+import numpy as np
+import cv2
+
+
+def get_image_feature(image_path, bins: tuple) -> tuple:
+    """
+    计算图像hsv特征值
+    :param image_path:
+    :param bins: HSV 所占比重
+    """
+    image = cv2.imread(image_path)
+    image_uid = image_path[image_path.rfind("/") + 1:][:6]
+    image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
+    # 获取图像中心点, 切割图像
+    (h, w) = image.shape[:2]
+    (c_x, c_y) = (int(w * 0.5), int(h * 0.5))
+    segments = [
+        (0, c_x, 0, c_y), (c_x, w, 0, c_y),
+        (c_x, w, c_y, h), (0, c_x, c_y, h)
+    ]
+
+    # 绘制椭圆轮廊
+    (axes_x, axes_y) = (int(w * 0.75 / 2), int(h * 0.75 / 2))
+    ellipse_mask = np.zeros(image.shape[:2], dtype="uint8")
+    cv2.ellipse(
+        ellipse_mask, (c_x, c_y), (axes_x, axes_y),
+        0.0, 0.0, 360.0, (255, 255, 255), -1
+    )
+
+    features = []  # 图像特征值
+    features_extend = features.extend
+    for (start_x, end_x, start_y, end_y) in segments:
+        corner_mask = np.zeros(image.shape[:2], dtype="uint8")
+        cv2.rectangle(
+            corner_mask, (start_x, start_y),
+            (end_x, end_y), 255, -1
+        )
+        # 逆时针计算图像边角颜色直方图
+        corner_mask = cv2.subtract(corner_mask, ellipse_mask)
+        image_histogram = calculate_histogram(image, corner_mask, bins)
+        features_extend(image_histogram)
+    # 计算中心椭圆颜色直方图
+    image_histogram = calculate_histogram(image, ellipse_mask, bins)
+    features_extend(image_histogram)
+    convert_features = np.array(features).astype(float)
+    return image_uid, convert_features
+
+
+def calculate_histogram(image, mask, bins: tuple) -> list:
+    """
+    计算hsv颜色直方图
+    :param image: 输入图像
+    :param mask: 计算区域
+    :param bins: bins: HSV 所占比重
+    """
+
+    hist = cv2.calcHist(
+        [image], [0, 1, 2], mask, bins,
+        [0, 180, 0, 256, 0, 256]
+    )
+    # 颜色直方图归一化
+    histogram = cv2.normalize(hist, hist).flatten()
+    return histogram

hsv_image_search/generate_index.py

@@ -0,0 +1,21 @@
+#!/usr/bin/env python
+# coding: utf-8
+
+import numpy as np
+
+
+def chi_square(query_feature, match_feature) -> float:
+    """
+    卡方校验
+    :param query_feature: 搜索图像
+    :param match_feature: 匹配图像
+    :return: 卡方距离
+    """
+
+    distance = 0.5 * np.sum(
+        [((a - b) ** 2) / (a + b + 1e-10)
+         for (a, b) in zip(match_feature, query_feature)]
+    )
+    return float(distance)
+
+

ORB-imageSearch/mlibs/__init__.py hsv_image_search/mlibs/__init__.py

@@ -0,0 +1,10 @@
+Flask==1.0.2
+gunicorn==19.9.0
+Werkzeug==0.14.1
+async-timeout==3.0.1
+httptools==0.0.11
+MarkupSafe==1.0
+numpy==1.15.0
+opencv-python==3.4.2.17
+tinydb==3.11.1
+uvloop==0.11.1

hsv_image_search/mlibs/hsv_features.py

@@ -0,0 +1 @@
+

hsv_image_search/mlibs/image_match.py

@@ -0,0 +1,50 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <title>{{ site_name }}</title>
+</head>
+<style>
+    .footer {
+        left: 0; bottom: 0;
+        width: 100%; color: white; text-align: center;
+    }
+    .mousse {
+        background-image: url("/static/css/background.jpg");
+        -webkit-background-size: cover;
+        -moz-background-size: cover;
+        -o-background-size: cover;
+        background-size: cover;
+        color: #000000;
+    }
+</style>
+
+
+<body class="mousse">
+<h1 align="center">{{ site_name }}</h1>
+<header class="intro" align="center">
+    <div class="intro-body">
+        <div class="container">
+            <h2 style="color: #c1d0eaa8">原图</h2>
+            <img src="/static/uploads/{{ upload_image }}" style="height: 250px;">
+            <h2 style="color: #c1d0eaa8">搜索结果</h2>
+            <table width="100%" border="0" cellspacing="0" cellpadding="0">
+                {% for search_result in search_results %}
+                  <tr>
+                    <td align="center">卡方距离(越小越好):{{ search_result[1] }} </td>
+                  </tr>
+                  <tr>
+                    <td align="center"><img src="/static/dataset/{{ search_result[0] }}.png" width="300" height="300" /> </td>
+                  </tr>
+                {% endfor %}
+            </table>
+        </div>
+    </div>
+</header>
+
+
+<footer class="footer">
+    <p>Mousse by 2018-11-3 © IG NB</p>
+</footer>
+</body>
+</html>