Vous êtes sur la page 1sur 10

FACULTY OF ENGINEERING

LAB REPORT SUBMISSION COVER PAGE

ECE3086 MULTIMEDIA TECHNOLOGY AND


APPLICATION

TRIMESTER 1 SESSION 2016/2017

Student Name:
Student ID:
Degree Major:
Declaration of originality:
I declare that all sentences, results and data mentioned in this report are from my own
work. All work derived from other authors have been listed in the references. I understand
that failure to do this is considered plagiarism and will be penalized.
Note that collaboration and discussions in conducting the experiments are allowed but
copying and any act of cheating in the report, results and data are strictly prohibited

Student signature:

Experiment title: MTA1 Image Processing for Image Retrieval


Table/PC No.:
Date Submitted: ................
Lab Instructor Name: Verified:.
(Please get your lab instructor signature after they have verified your result)
ABSTRACT :

The objectives of this experiment is to get more understanding with the Octave software to apply
basic image processing methods using Octave. Besides that, this experiment also teach us to
perform image feature extraction and compare the feature effectiveness in differentiating
different categories of image. Lastly it isto perform comparative analysis of edge detection
algorithm.

INTRODUCTION :
The CBIR system that uses the QBE approach involve two main process that are the database
preparation and image searching In the database preparation process, every image in the database
will need to be indexed or represented by a compact image descriptor (feature vector). The
annotation is done automatically whereby the CBIR system extracts and stores the feature vector
from the database image. In the searching process, the CBIR search engine basically needs to
extract the feature vector from the query image and compare this feature vector with the feature
vectors of the images in the database.
A better approach is to automatically extract feature vector from the image and use the feature
vector as a searchable index used by the image search engine. Feature vector describes the
Attributes of the object is described by the feature vector. The feature vector encodes the
measurements obtained from the various attributes of the image. Feature vector is used to
categorize objects or images into groups. It serves as a compact image descriptor. The CBIR
system will retrieve a set of similar images to the query image. In order to let us retrieve the
similar images, we need to use the retrieval algorithm to compare the dissimilarity between the
query image and the images in the database.
The Euclidian distance between the feature vector H1 and H2 is denoted as D E(H1 , H2 )

The larger Euclidian distance between the image descriptor H1 and H2 means the more dissimilar the
image I1 when compared to I2
METHOD :

1. Files from MMLS given by the lecturer was loaded into the folder
C:\MATLABR11\work\cbir.

2. Wang dataset was used which as been manually annotated. The prepared database has been
saved in the mat file database_cbir_2D_hue_value.mat. The database was loaded into Matlab
workspace after entering the working directory. The given worksheet was followed as reference.
cd('C:\MATLABR11\work\cbir')
load database_cbir_2D_hue_value.mat
The database contains the 2-dimensional histogram of the hue and value channels. The code used
to prepare the database is demo_create Database_all.m . The code can be modified to extract
different type of feature vectors.

3. 11 bins were used for each color channel to construct the histogram as the image descriptor.

4. 10 images was selected from each category as the query images. For each query images, 10
most similar images was found with the use of euclidian distance and the colour histogram
feature. The average precision rate was computed for all the retrieval attempts. Result was
tabulated

5. A few retrieval results was included in the report.

RESULT:

2D Colour Histogram 3DColour Histogram


Category Query Image
Precision Recall Precision Recall
Rate, PR Rate, Rate, PR Rate,
RR RR
Beach 100.jpg 2 2 2 2
10 = 0.2 90 = 10 = 0.2 90

0.022 =0.022

Building 203.jpg 5 5 10 10
10 = 0.5 90 = 10 = 1.0 90 =

0.056 0.111
Bus 306.jpg 10 10 10 10
10 = 1.0 90 = 10 = 1.0 90 =

0.111 0.111

Dinosaur 403.jpg 10 10 10 10
10 = 1.0 90 = 10 = 1.0 90 =

0.111 0.111

Elephant 508.jpg 5 5 3 3
10 = 0.5 90 = 10 = 0.3 90 =

0.056 0.033

Flower 603.jpg 8 8 8 8
10 = 0.8 90 = 10 = 0.8 90 =

0.089 0.089

709.jpg 4 4 5 5
House 10 = 0.4 90 = 10 = 0.5 90 =

0.044 0.056

801.jpj 5 5 5 5
Mountain 10 = 0.5 90 = 10 = 0.4 90 =

0.056 0.056

900.jpg 8 8 10 10
Food 10 = 0.8 90 = 10 = 1.0 90 =

0.089 0.111

TABLE 1
Precision rate PR = CI/NR
Recall rate RR = CI/NC

CONCLUSION :

In conclusion, from this experiment I had learned how to use the octave software. Besides I learn
how to retrieve an image from image processing.

Vous aimerez peut-être aussi