Matching and Retrieval System Based on Vocabulary and Grammar of Color Patterns By

Download Report

Transcript Matching and Retrieval System Based on Vocabulary and Grammar of Color Patterns By

Matching and Retrieval System
Based on Vocabulary and
Grammar of Color Patterns
By
Pradeep C.Venkat
Srinath Srinivasan
The Problem



To design an intelligent perception based
system for pattern matching and retrieval
of patterns from a database
System must retrieve the closest match(es)
in terms of ‘similarity’ to a user-inputted
query image
Matching must be done so as to emulate
human perception to the extent possible
Mimicking Humans is a Tough Job!



How do humans judge ‘similarity’ of
images?
What factors would best characterize the
subjective phenomenon of human
perception?
Can these factors be generalized over all
kinds of images?
Our Approach - Vocabulary and Grammar





o
Vocabulary:
Four perceptual criteria (Mojsilovic, et al.) were
identified for comparison of color patterns:
Overall color
Color purity
Regularity and Placement
Directionality
Grammar:
A set of rules governing the use of these criteria in
judging similarity of patterns
Overview of the System
Image
Decomposition
Feature Extraction
Generation of
Pattern Map
Image
Database
query
Extraction of
Texture
Primitives
Estimation of
Primitive
Distribution
Similarity
Judgment
Similarity Measurement
Estimation of
Color
Distribution
Steps for Feature Extraction (Color Based)




The input image is transformed into the Lab Color
Space for compact perceptually based color
representation
Color distribution is determined using a vector
quantization approach
Significant features are determined from the
histogram
Color features are used in conjunction with an L2norm distance measure to determine similarity
Steps for Feature Extraction (Texture Based)

Spatial smoothing to remove background noise

Construction of the achromatic pattern map (APM)

Construction of an edge map from the APM


Orientation processing to extract the distribution of
pattern contours along different spatial directions
Computation of scale-spatial texture edge
distribution
The Database

Our database consisted of over 300 images
of color patterns, sceneries, buildings,
plants, etc.
Results (Obvious Matches)
Query
Closest Matches
Results (contd.) - Obvious Matches
Query
Closest Matches
Results contd. (Non Obvious Matches?)
Query
Closest Matches
References


A. Mojsilovic, et. al, Matching and retrieval based on vocabulary
and grammar of color patterns, IEEE Trans. Image Processing,
vol. 9, no. 1 (Special issue on Digital Libraries), Jan. 2000, pp.3854.
A. Mojsilovic, et. al, The vocabulary and Grammar of Color
Patterns, IEEE Trans. Image Processing, vol. 9, no. 3 (Special
issue on Digital Libraries), Jan. 2000, pp.38-54.