An Introduction To

The Problem of Finding Images in Large Databases

A Technical Seminar

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

            

Existing System(TBIR) and Drawbacks Why CBIR? Typical CBIR System CBIR Query Feature Extraction Content Comparing & Clustering Use of Relevance Feedback Sample Algorithms(Edge Detection & Texture Formation) Use of Segmentation & Sample Algorithm CBIR Drawbacks Applications Real-Time CBIR Systems References
Department Of Computer Science & Engg

PSG College Of Technology, Coimbatore

Keywor ds Query Who are the Staffs of PSG?

Who, staffs, PSG

Imag e Set 1 Who, staffs, PSG Imag e Set 2 Staffs, PSG Result Set (As Per Ranking)

Search Engine

Conventional Database [Images + Description]

: :

Use r

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

 

It is not possible to enter all descriptions of a single image. If the Image contains Geographical Data then the manual Keyword descriptions are not possible. Ex: Google Earth contains Huge Amount of Geographical Data. It is Cultural Language Dependent. It is not possible to describe every image in the Database. Keywords matching will not give the most relevant images.

  

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

CBIR is originate from fields such as statistics, pattern recognition, signal processing and Image Processing.

 "Content-based" means that the search will analyze the actual contents of the image.  The term 'content' in this context might refer to colors, shapes, textures, or any other information that can be derived from the image itself.

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

1.

Query by example • Select existing sample Image from DB as Query image. • Draw a sketch of the image . Semantic retrieval • More complex Text Queries, Like “List all dogs look like Cat” Others • Browsing in a desired Image Category. • Querying by image region (rather than the entire image). • Querying by multiple example images.

1.

1.

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

FV

FV

...
Search by: •Examples •Sketches •Text Queries

Visual features

Database

CBIR Search Engine

Relevance Feedback

Clustering & Categorizing

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

  a 

Feature is the Representation of Specific Structure of Image One Feature may not give better results, so find out more Features for single image for better Relevant Results. So the features of a specific Image is called as Feature Vector. Transforming the input image data into the set of Feature Vectors is called features extraction.
Color Histogram

Image

Shapes Textures

Features Vector

Database

.. .
PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

1.

Color • Retrieving images based on color similarity is achieved by computing a color histogram for each image. • Ex. Color Histogram Shape • Shape does not refer to the shape of an image but to the shape of a particular region. • Ex. Segmentation & Edge-Detection. Texture • Texture is usually defined as a pattern that repeats itself regularly in the image • Ex. Gabor Filters

1.

1.

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

Developed By G.BAKTHAVATCHALAM
For(I=1 to width-1) For(J=1 to height-1) R = Red(Pel(I - 1, J - 1) Xor (Pel(I, J)) Or 1 G = Green(Pel(I - 1, J - 1) Xor (Pel(I, J)) Or 1 B = Blue(Pel(I - 1, J - 1) Xor (Pel(I, J)) Or 1 If R > 255 Then R = 255 If G > 255 Then G = 255 If B > 255 Then B = 255 FeatureVector(edge, I, J, Pel(R, G, B))

i-1,j-1 i-1,j

i,j-1 i,j

i+1,j-1 i+1,j

i-1,j+1 i,j+1 i+1,j+ 1

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

Developed By G.BAKTHAVATCHALAM
For(I=1 to width-1) For(J=1 to height-1) R = Red (Pel(I - 1, J - 1) - (Pel(I, J)) +128 G = Green (Pel(I - 1, J - 1) - (Pel(I, J)) +128 B = Blue (Pel(I - 1, J - 1) - (Pel(I, J)) +128 NO = Abs((R + G + B) / 3) If NO > 255 Then NO = 255 FeatureVector(texture, I, J, Pel(NO, NO, NO))

i-1,j-1 i-1,j

i,j-1 i,j

i+1,j-1 i+1,j

i-1,j+1 i,j+1 i+1,j+ 1

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

Image

.. .
Comparison (Nearest Neighbors)

Image

Image Feature Vectors

Image Feature Vectors

Clusters / Categories

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

• Divide the image into several segments(regions) and color or texture of each region is approximately uniform, – Describe each region. – Measure similarity between regions. – Combine information about similarity between regions into a similarity measure between images.

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

Developed By G.BAKTHAVATCHALAM
For(I=1 to width-1) For(J=1 to height-1) R = Red(Pel(I + 1, J) Or (Pel(I, J)) Xor 1 G = Green(Pel(I + 1, J) Or (Pel(I, J)) Xor 1 B = Blue(Pel(I + 1, J) Or (Pel(I, J)) Xor 1 If R > 255 Then R = 255 If G > 255 Then G = 255 If B > 255 Then B = 255 FeatureVector(seg,I, J, Pel(R, G, B))

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

1. 2.

After retrieval of results, the irrelevant images also may found. User give their feedback about the results to avoid images that are not related to the given query.
• • • Relevant Not-Relevant No-Idea

1.

After that the image which are irrelevant is not included for that query in future.
Database Images Feature Extraction

Lookup

Feedback

Feedback

Query Image Result
PSG College Of Technology, Coimbatore

Better Result

Better Result
Department Of Computer Science & Engg

 More Resources are needed like CPU time, memory, disk space, etc.  Determining Semantic meaning of Objects is not possible.  Results may contain False Hits.  Problems on Processing Different aspects of Images.

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

1. Image Search Engines 2. Investigations • Compare Images • Collect Relevant Images 1. Security • Thumb Recognition 1. Analysis • Geographical • Enterprise Related 5. Medical diagnosis 6. The Military 7. Remote sensing systems

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

1. Viper (http://viper.unige.ch/) Developed by the University of Geneva Hospitals, Viper stands for Visual Information Processing for Enhanced Retrieval. 2. GIFT (http://www.gnu.org/software/gift/) It is a free GNU image finding tool developed by the Viper project. 3. QBIC (http://wwwqbic.almaden.ibm.com/) IBM’s Query by Image Content allowing one to query large image databases based on visual image content. 4. Photobook (http://vismod.media.mit.edu/vismod/demos/facerec/basic.html) MIT’s Photobook project is able to perform object and facial recognition by analyzing and comparing the individual components of an image. 5. Berkeley’s Digital Library Project (http://elib.cs.berkeley.edu/) Computer vision meets digital libraries – the digital library project contains a specific analysis of Image Retrieval by Image Content.
PSG College Of Technology, Coimbatore Department Of Computer Science & Engg

• CLUE: Cluster-Based Retrieval of Images by Unsupervised Learning By Yixin Chen, Member, IEEE, James Z. Wang, Member, IEEE, and Robert Krovetz. • OSDB2_PostGIS_Presentation by David Blasby Refractions Research dblasby@refractions.net , http://postgis.refractions.net • A. W. M. Smeulders, M. Worring, S. Santini, A. Gupta, and R. Jain, “Content-Based Image Retrieval At The End Of The Early Years” IEEE Trans. Pattern Anal. Mach. Intell., vol. 22, no. 12, pp. 1349–1380, Dec. 2000. • J. R. Smith and S.-F. Chang, “VisualSEEK: A Fully Automated Content Based query system” in Proc. 4th ACM Int. Conf. Multimedia, 1996, pp. 87–98.

PSG College Of Technology, Coimbatore

Department Of Computer Science & Engg

Sign up to vote on this title
UsefulNot useful