Content Based Image Retrieval

​ Xiaohe Xue Jan 14th, 2016

Introduction

​ This project is a rewritten version of Searching Images project in Chapter 7 in the book-Programming Computer Vision. This project has a simple but integrated ability of content based image retrieval. This project has five parts, creating vocabulary, indexing the image, searching in the database, ranking the images and using in web demo.

Project Structure

###Creating Vocabulary (Vocabulary)

• Property

  attribute	Type	DECR
  name	string	name of vocabulary
  voc	list	visual words, content of voc
  idf	list	idf value of every visual word
  trainingdata	list	path of sift file of every image
  nor_word	int	number of visual words

• Function

  1. train(self, featurefiles, k, subsampling)

     to create vocabulary. First read all the sift files of every image. Second, calculate the visual words by using k-means clustering.

     Finally, go through all training images and project on vocabulary.

     ​

  2. project(self, descriptors)

     to create the histogram of every visual word

     ​

Indexing the image (Indexer)

• Database Tables

  • imlist

    type	name	descr
    string	file	image path

  • imwords

    type	name	descr
    int	imid	image id
    int	wordid	visual id
    string	vocname	vocabulary name

  • imhistograms

    type	name	descr
    int	imid	image id
    histogram	pickle	histogram value in pickle
    vocname	string	voc name

Ranking

​ show two different ways to rank the images, one is regular way, another is in homography way

​ the ranking functions need to be improved

Search the images

​ use basic inverted list which use visual words as primary element to implete searching the image retrieval.

Implements in web

​ use flask web framework

Notes

1. need to use tf-idf

Requirements

1. PCV

2. pysqlite2

3. pickle

4. PIL

5. numpy

6. pylab

7. scipy

8. vlfeat commad lines (shell not python modlues)

   http://www.vlfeat.org/install-shell.html

9. sqlite (databse)