An efficiently phase-shift frequency domain method for super-resolution image processing

An Efficiently Phase-Shift Frequency Domain

Method for Super-Resolution Image Processing

Cao Bui-Thu

Department of Electronic Technology

Ho Chi Minh City University of Industry (HUI)

Ho Chi Minh City, Vietnam

e-mail: [email protected]

Tuan Do-Hong

Division of Electronic-Telecommunication

Ho Chi Minh University of Technology

Ho Chi Minh City, Vietnam

e-mail: [email protected]

Thuong Le-Tien

Department of Electric-Electronic Engineering

Ho Chi Minh University of Technology (HCMUT)

Ho Chi Minh City, Vietnam

e-mail: [email protected]

Hoang Nguyen-Duc

Science and Technology Research Center BARC

Vietnam Television VTV

Ho Chi Minh City, Vietnam

e-mail: [email protected]

Abstract— How to reconstruct a high resolution and quality from

low resolution images captured from a digital camera is always

the top target of any image processing system. Exploit the

aliasing feature of sampled images, we propose a new technique

to register exactly the motions between images, including

rotations and shifts, by using only frequency domain phase-shift

method. Based on registered parameters, the precise alignment of

input images are done to create a high-resolution by using

interpolation methods. It is more exactly when we compare our

algorithm to other algorithms in simulation and practical

experiments. The visual results of super-resolution images

reconstruced by our algorithm are better than that of other

algorithms. It is possible to apply our algorithm to increase

resolution of digital camera or video systems.

I. INTRODUCTION

In fact that quality of images from digital cameras is not

higher up to now. The reason for that is the limited resolution,

optical blur, vibration or motion of camera, aliasing in

frequency domain of pictures and noise.

Moreover quality of an image is proportion of the

resolution and shape of image details. So to increase quality of

image, we have to increase the resolution and fidelity of image

details.

The main purpose of our article is exploit the aliasing in

frequency domain of images sequence which due to vibration

or motion when captured from low resolution camera, to

develop a new method for reconstructing a high-resolution

image and increasing fidelity of image details.

From images sequence, captured on the same scene with a

lightly moving of the low resolution camera, the first frame of

image is selected as the main frame and the other are reference

frames. Basically, all of them are different slightly from the

other by shift, in vertical and horizontal, and rotation. Hence,

there are two main steps for super-resolution image

reconstruction. First we have to register exactly shifts and

rotations between main frame and the other. This is a vast

different challenge because a small error in the motion

estimation will translate almost directly into large degradation

of the resulting high resolution. Second we rearrange them in

the same coordinate, then use interpolation methods to

combinate the detail inform all of images to reconstruct and

create a high resolution image.

Super-resolution image reconstruction was first set up by

Tsai and Huang [2] in 1984. They described an algorithm to

register multi-frame simultaneously using nonlinear

minimization method in frequency domain. Their algorithm

had not good result because of aliasing images in frequency

domain. Up to now, there are many authors and their methods

for super-resolution image reconstruction described in technical

overview of Park [1] in 2003.

In general, there are two main approach of super-resolution

reconstruction. The first is frequency domain approach. Most

of frequency domain registration methods are based on the fact

that two shifted images differ in frequencey domain only by a

phase shift, which can be found from their correlation in

Fourier transform. Reddy and Chatterji [8] in 1996, apply a

high-pass emphasis filter to strengthen high frequency for

estimating motion. Stone et al [9] in 2001, also applied a phase

correlation technique to estimate planar motion. Lucchese and

Cortelazzo [6] in 2000 developed a rotation estimation

algorithm base on the property that the magnitude of Fourier

transform of an image and the mirrored version of the

magnitude of the Fourier transform of a rotated image has a

pair of orthogonal zero-crossing lines. The angle that these

lines make with the axes is equal to haft the rotation angle

between images. Latterly Vandewalle [4] in 2006 develop a

rotation estimation algorithm by dividing image into small

segment from the centre of the images, then using phase

correlation in power spectrum of each angle segment.

The 2009 International Conference on Advanced Technologies for Communications 978-1-4244-5139-5/09/$26.00 ©2009 IEEE