content based image retrieval using color texture

7/28/2019 Content Based Image Retrieval Using Color Texture

1/19

A

Presentation on

Content Based Image Retrieval

Using Color, Texture and Shape FeaturesSeminar II

- Presented by -

Sachin Deshmukh (M.E. Computer)99406068

- Guide -

Amrutvahini College of Engineering, Sangamner


2/19

Outline

Introduction

How it works

Characteristics of image queries

Feature Extraction

- Color Feature Extraction- Texture Feature Extraction

- Shape Feature Extraction

Indexing and retrieval

NOHIS Search tree

Conclusion


3/19

Introduction

What is content based?

What is CBIR?

Why CBIR?

- Digital image database growing rapidly.

- Professional needs.- Difficulty in locating images on the web.


4/19

How it works

What is content based?

Query Image Extract Features

Image Database Feature Database

Similarity Measure Feedback Algorithm

Final matched result


5/19

Characteristics of image queries

Retrieval by primitive features such as color, texture, shape

or the spatial location of image elements.

Retrieval by derived (sometimes known as logical) features,

involving some degree of logical inference about the identityof the objects .

retrieval of objects of a given type

retrieval of individual objects or persons

Retrieval by abstract attributes, Significant amount of high-

level reasoning about the meaning and purpose of the objects

or scenes.


6/19

Feature Extraction

In CBIR, each image that is stored in the database has its

features extracted such as color, texture, shape and compared

to the features of the query image.

Some Feature Extraction techniques are - Color Feature Extraction

- Texture Feature Extraction

- Shape Feature Extraction


7/19

Color Feature Extraction

DCT Coefficient Transformation

The discrete cosine transform is a fast transform. It has excellent

compaction for highly correlated data. By using DCT, minimize the

number of bits required to represent the information in an image, by

removing the redundancy between neighboring pixel values.

Where,

- u is the horizontal spatial frequency,- v is the vertical spatial frequency,

- f(x, y) is the pixel value at coordinates (x, y),

- C(u, v)is the DCT coefficient at coordinates(u, v).


8/19

YcbCr represents color as brightness and two color difference signals.

In YCbCr, the Y is the brightness (luma), Cb is blue minus luma (B-Y)

and Cr is red minus luma (R-Y). The transformation of RGB to YCbCr

color model can be derived as a linear transformation

Y = 0.299R + 0.587G + 0.114B

Cb = -0.1687R - 0.3313G + 0.5B

Cr = 0.5R - 0.4187G - 0.0813B

In order to reduce the computational load and it decouples intensity

and color information this conversion is done.

DCT transformed image is quantized by reducing the high frequency

pixels simply by dividing it by a constant quantization matrix andthen rounding it. A lot of the values will become too small. All the

quantized DCT coefficients are stored in an array of feature vector.

Color Feature Extraction


9/19

The images are segmented, into sub image. Each sub image is

transformed using a 2D-DFT transformation. After DCT only

magnitude coefficient matrices are used for texture

characterization. All the 128x128-pixel sub images obtained

from image segmentation are transformed via a 2D-DFTaccording to equation.

Where,

- u is the horizontal spatial frequency,

- v is the vertical spatial frequency,

- f(x, y) is the pixel value at coordinates (x, y),

- (u, v) is the DCT coefficient at coordinates(u, v).

Texture Feature Extraction


10/19

Shape representations can be generally divided into two

categories:

Boundary based

Region based

Boundary-based shape

representation only uses the

outer boundary of the shape,

i.e., the pixels along the object boundary.

Region-based shape representation uses the entire shaperegion by describing the considered region using its internal

characteristics; i.e., the pixels contained in that region.

Shape Feature Extraction


11/19

Indexing is a kind of sorting based on the value given to the

image after finding the similarity of each images. It is used to

accelerate the query performance in the search process and

plays a main role in supporting effective retrieval of sequences

of image. Proper indexing makes the search easy and efficient.

Indexing using Hierarchical Grid Based Technique

In this approach the image is converted into the YCbCr color

space and the feature vector derived from the discrete cosine

transform coefficients through vector quantization, eachimage corresponds to a grid code and visually similar patterns.

Indexing


12/19

indexing techniques

- The data-partitioning

- The space-partitioning

When the nearest neighbors search is applied on a data

partitioning index, additional clusters are visited due to theoverlapping between the bounding forms. By using NOHIS

(Non Overlapping Hierarchical Index Structure) tree the

overlapping is avoided and the quality of clusters is preserved

Indexing using NOHIS tree


13/19

1) The entire set of descriptors extracted from the image

database. This cluster is divided into two sub-clusters using the

hierarchical clustering algorithm PDDP.

Each of the two sub clusters is divided into two partitions

recursively and arranged into binary tree.

2) Descriptors of each obtained sub-clusters are gathered

by hyper-rectangles directed according to the leading

principal component to ensure the non-overlapping between

the two bounding forms.

Indexing using NOHIS tree


14/19

K-NN Search (K Nearest Neighbors search)


15/19

K-NN Search (K Nearest Neighbors search)


16/19

Some comparative work


17/19

Some comparative work


18/19

Conclusion

More image databases, embracing color images, with

different textures, produced under varying lighting conditions

we need to develop more intelligent CBIR system.

The performance evaluation of the proposed system

with other systems shows that NOHIS-Search is faster that thetwo other systems. NOHIS-Search, however, requires further

investigations especially in the matching process.

By using Techniques like NOHIS search we can able to

implement efficient CBIR system.


19/19

THANK YOU!

content based image retrieval using color texture

Documents