8/6/2019 jcirimele

1/2

My Wonderful Article in IEEE Format

Jesse E Cirimele

Department of Cognitive Science

UCSD

San Diego, CA

Abstract

This project will be my implementation of a super resolution

algorithm. It will first be tested using a text recognition

program and some toy data. Then it will be tested under the

specifications of a vision assistant.

1. Introduction

There have been several methods of super resolution for

digital images developed. Some involve hallucinating ex-

tra detail based on similar types of images, others involve

using multiple low resolution images to get one high reso-

lution image. I will be implementing an algorithm of the

later type.

I will be testing the practical effectiveness of the imple-

mented algorithm using an OCR (optical text recognition)

program on first the low resolution images and then the re-

sulting high resolution image. Objective improvement inimage resolution can then be determined on the percent of

correct identification of words.

My final step will be to prepare the algorithm for use

with a vision assistant. In order to do this I will have to

re-implement the algorithm in C++ and test it for images

that conform to the type and resolution that will be acquired

from the device.

2. Project Details

2.1. AlgorithmI will be using a super resolution algorithm discussed by As-

saf Zomet and Shmuel Peleg [1]. This method starts off by

modeling the process of imaging. This includes modeling of

geometric transformation, blurring, subsampling and noise.

The super resolution can be calculated using a conjugate-

gradient based approach [1]. Zomet and Pelegs approach

references work by Elad and Feuer, 1999, Capel and Zisser-

man, 1998, Capel and Zisserman, 2000.

2.2. Data Set

For my first data set I plan on using images from

a video sequence used by Peleg and Zomet. This

video sequence can be downloaded from this webpage:

http://www.cs.huji.ac.il/ zomet/superResolution.html

For my second data set I will acquire my own images of

street signs and bus signs which will be similar to those thatmight be acquired and used by the vision assistant.

2.3. Test Method

One pretty objective way to test whether the algorithm

created improvement would be by using a text recogni-

tion system. I found simpleocr which is a free opti-

cal character recognition software. it can be found at

http://www.simpleocr.com/. Using this program on the high

resolution image if available, the low resolution images, and

the high resolution image generated from the super resolu-

tion algorithm should give tangible evidence on the effec-

tiveness of the super resolution implementation.

2.4. Timeline

Here is a rough timeline for the project

Week 1: read paper thoroughly

Weeks 2-3: implement algorithm

Week 4: test on video sequence data set

Week 5: collect data set 2

Weeks 6-7: use algorithm on second data set and test

with simpleocr

Weeks 8-9: Implement in C++

Weeks 9-10: write up of research

1

8/6/2019 jcirimele

2/2

References

[1] A. Zomet and S. Peleg, Super Resolution from Multiple

Images Having Arbitrary Mutual Motion, in S. Chaudhuri

(Editor), Super-Resolution Imaging, Kluwer Academic, Sept.

2001, pp. 195-209.

[2] A. Zomet, A. Rav-Acha and S. Peleg, Robust Super Resolu-tion, Proceedings of the International Conference on Com-

puter Vision and Pattern Recognition (CVPR), Hawaii, De-

cember 2001.

[3] D. Capel and A. Zisserman, Automated Mosaicing with

Super-Resolution Zoom, IProc. CVPR, pages 885891, Jun

1998.

[4] D. Capel and A. Zisserman, Super-resolution enhancement

of text image sequences, Proceedings of the International

Conference on Pattern Recognition, 2000.

2