In order to incrementally update tag-tag and image-image relations, we implement an efficient function to modify related tables.

Usage Guide:call the function: Add_NewTag($imageID, $flickrID, $tag)
(Please include the file "common/AddNewTag.php" first.)
You have to provide $imageID or $flickrID.
Besides, the name of added $tag is also needed.

Seven tables will be updated when one tag is added into an image.
Tag related tables:
* TBL_ImageTag
* TBL_InvertedFile
* TBL_TagTagMatrix
* TBL_TagRelationSVD

Image related tables:
* TBL_InvertedFileTrans
* TBL_ImageImageMatrix
* TBL_ImageRelation

All tables can be updated within 1~2 seconds.

- Chi-Yao

Rent Progress

Following items are finished.

Tag-Tag Relation
Image-Image Relation

Incremental Upate of adding new tags will be done soon.

Singular Value Decomposition

Reference Website: http://www.alglib.net/matrixops/general/svd.php

Final Image Categories in the Database

each type 500 images:

1. scenery
2. landscape
3. landmark
4. night

5. street
6. building
7. river
8. lake
9. beach
10. mountain
11. forest
12. sunset
13. sunrise
14. sky

The first 4 categories are somewhat general.

Problem decomposition

Milestone 1:
1. Given a tag, retrieve images by ranking.
2. Given a tag, retrieve tags by ranking.
3. Given multiple images, retrieve images.
4. Given multiple images, retrieve tags.
5. Given an image, retrieve images.

Computing Eigenvalues and Eigenvectors

Gansterer, W; Schneid, J; Ueberhuber, C.; "A Low-Complexity Divide-and-Conquer Method for Computing Eigenvalues and Eigenvectors of Symmetric Band Matrices"; Bit Numerical Mathematics, Volume 41, Number 5, December 2001 , pp. 967-976(10)

http://www.ingentaconnect.com/content/klu/bitn/2001/00000041/00000005/00331161

IBM MARVel: Multimedia Search Engine

Hey guys,
IBM has built an amazing multimedia search engine.
Please check this site.
MARVel: Multimedia Search Engine

Incremental IDF

According to today's discussion, we will use tf-idf model to rank the relevance values of images. However, since our tags are incremental, the idf values are changed over time. I found some related works on this issue.
FYI.

Anthony Tomasic, Hector GarciaMolina, and Kurt Shoens, "Incremental Updates of Inverted Lists for Text Document Retrieval," In Proceedings of the 1994 ACM SIGMOD Conference, pages 289--300, New York, USA, 1994.

Sergey Melnik, Sriram Raghavan, Beverly Yang, and Hector Garcia-Molina, "Building a Distributed Full-Text Index for theWeb," In 10th World Wide Web, pages 396--406, 2001

L. Lim, M. Wang, S. Padmanabhan, J. S. Vitter, and R. Agarwal, "Dynamic Maintenance of Web Indexes Using Landmarks," In Proceedings of the 12th International World Wide Web Conference, pages 102--111, May 2003.

TRECVID 2003 Collection

http://www-nlpir.nist.gov/projects/tv2003/tv2003.html


I find that there are several works using TRECVID 2003 collection as their evaluation dataset. TRECVID stands for TREC Video Retrieval Evaluation and is used to evaluate content-based reterival algorithms. TRECVID defines 17 different categories. Please take a look at the above link when you are free. :)

Interesting problems

On Nov 15, we discussed more details about the project's goal and implementation.
It seems to us that there are (at least) three interesting problems listed as below:
1) Tag seach and recommendation by incorporating association rules
2) Image search by multiple query images
3) Visualization model

Our first step is to construct the initial image database and find out the possible solution for problem 1.

Testing Enviroment

I've set up a testing environment to build our system.
The OS is Ubuntu v.7
Apache2, mysql, and php are already installed.
I will email the id and password to you guys.
If there is any additional requirement, please feel free to let me know.
BTW, the machine is shared with other members in our lab.
Maybe I could find another one to be used by ourselves. :D

Flicker API: Download Photos

Perhaps there has been better way to download photos, the reference below shows the sample code for getting the url of targeted photos.

References:
http://phpflickr.com/example-a-users-photos/?source

Flicker API: search and extract tags

Sample code:
require_once("phpFlickr.php");
$f = new phpFlickr("api_key");

echo "--------------photos_search----------------
";
$search_args = array("text"=>"sunset","sort"=>"relevance");
$search_info = $f->photos_search($search_args);
$i = 1;
foreach ($search_info['photo'] as $photo)
{
$owner = $f->people_getInfo($photo['owner']);
$photo = $f->photos_getInfo($photo['id']);
echo "photo_id: ".$photo['id']."
";
echo "title: ".$photo['title']."
";
echo "tag[0]: ".$photo['tags']['tag'][0]['raw']."
";
if($i==3) break; $i++;
}
echo "==========================================";


Readme for search function:
/* This function strays from the method of arguments that I've
* used in the other functions for the fact that there are just
* so many arguments to this API method. What you'll need to do
* is pass an associative array to the function containing the
* arguments you want to pass to the API. For example:
* $photos = $f->photos_search(array("tags"=>"brown,cow", "tag_mode"=>"any"));
* This will return photos tagged with either "brown" or "cow"
* or both. See the API documentation (link below) for a full
* list of arguments.
*/

References:
http://flickr.com/services/api/

PHP executes C++ code

Hi All:
After I do the brief survey, it seems that to execute c++ code in php is viable.
Maybe the c++ code has to be compiled in UNIX environment rather than in windows.
Moreover, to link mysql database in c++ code is also viable.
I have implemented it in the project of one of my previous course.

Functions: exec() or passthru()
References:
http://tw2.php.net/passthru
http://digi.it.sohu.com/20050421/n240014600.shtml

Regular discussion

On Nov 8, we discuss the development environments and its related technical issues. Jen-Wei propose using Java which is cross-platform. And Chi-Yao suggest using PHP to show the web interface.
The evaluation method will be the critical path, and one of the possibility is to compare with ordinary one. However, we prefer to calculate the precision-recall values for each steps. Therefore, we should get higher score after more trials.

Flicker API

Hello,
I have investigated Flicker API and tested the PHP API recently. It is fortunate that it really works. The next step of our project may be to design implementation details. Good luck for us.
- Tonnny

MyHeritage

Hey guys, I found sth. very interesting.
http://www.myheritage.com

The above website allows you to upload your own photo.
This site is able to automatically recognize your face in the uploaded photo.
In addition, it matches your face to famous celebrities and provides a list of matching results.
http://www.myheritage.com/FP/Company/face-recognition.php

Moreover, you can create your own family tree on this website
and share your photos to your friends in the world.

Check it out and enjoy the list. :P
(There is a version in simplified Chinese.)

Here are some webpages introducing MyHeritage website.
http://blog.roodo.com/hope01/archives/2831857.html
http://www.wretch.cc/album/album.php?id=ariel816&book=10&page=1

3rd meeting

Time: Oct. 22
Venue: NetDB Lab
We hold a meeting this time in order to see another proposl possibility. But we don't have a big progress. Therefore after addressing some issues for the PIXAR, we decide to tell Prof. Hsu about our idea (PIXAR). We need his feedback. So Jen-Wei will arrange the schedule with Prof. Hsu.

2nd meeting

Time: Oct. 18
Venue: NetDB Lab
This time we discuss the proposal proposed by Jen-Wei. He names the name as PIXAR (Progressive Image eXtraction And Recommendation).
The goal of this term is mining the user's behavior not only from the static tag information but also the navigation data.
Three phrases are included in this implementation: query, recommendation, and relation.
From the time and resource perspective, this term is not a small-scale project. Therefore, we plan to leverage some existing platform such as Flickr. The Flickr provides some open APIs that can support image query.
Chi-Yao also proposes his idea about the persoanl photo management. For example, filtering out some photos that the visual quality is not good. We suggest Chi-Yao to write down more specific research tasks next time to share with us.

startup

Oct 11:
We discuss the potential area for our term project. Basically, we will focus on 'image' media.