SUI Shou-xin

（Ocean University of China,Qingdao Shandong 266100）

0 Introduction

High Dynamic Range Imaging is applied to the fields of computer graphics,visual information acquisition and processing and creative digital media.It’s a new imaging technology developed in these decades.It can achieve more exposure dynamic range than ordinary digital image technology，and can truly express the actual world.

In the next few years,high dynamic range imaging technology will show more and more important value in academic research,family life,studio entertainment,industrial application,etc.But,from Photomatix the desktop application software for still images to the high definition wide dynamic camera for the video in the industrial application,and then to high-end audio and video consumer electronics of iphone4.there is a series of common problems that the video images are easy to fuzzy and double in case of movement.However,from all of the literatures published in recent years,we still can’t find any method which is able to solve these problems.In order to capture the future,now it is necessary to have a deep and detailed study on a series of basic science and the core technology of high dynamic range imaging,and pay more attention to the unsettled problems from Market feedback.

There are two ways to acquire HDR image.One is using ordinary camera based on different exposure to get more images of the same scene;Another is directly outputting 10 bits of data from the CMOS or CCD image sensor,and this image is also called the Raw image.De Neve scholar put forward a CRFS and RM combined calculation and optimization method[1].Kartalov proposed a mathod that performing perpixel synthesis for two pieces of different exposure Images in the spatial domain[2];Wu Xiaojun presented a HDRI generating method based on image segmentation and merging[3];Li Zhengguo put forward a two-way prediction method to solve the double problem that appear in HDR image synthesis[4].From these literatures we can see,in the HDR image synthesis,the common problem that still need to be sloved is how to merge many images efficiently and directly to the high dynamic range image which can be displayed,And these algorithms above are complex to get the clear HDR fusion image.The method in this paper is relatively simple and can get the detail information of the image.It is a new choice in High Dynamic Range Imaging processing.

This article mainly aims at solving the high dynamic problems above,and proposes a high dynamic image fusion technology based on wavelet transform.The rest of this paper is organized as follows,section II introduces the thoery of wavelet transform especially the haar wavelet decomposition.section III proposes the realization of this algorithm and analyzes the results by comparing the processed images with the results using the software of Photomatix.Finally in section IV shows the conclusion and summarizes the significance of this paper.

1 Wavelet transform

Wavelet transform is a new mathematical tool,its inherent properties of multi-scale and multi-resolution make it shows the following advantages in image processing:Perfect reconstruction ability of wavelet transformation ensures that the signals will not loss any information in the process of decomposition,and also won't produce any redundant information. Wavelet transform can decompose the image into approximation image and detailed image,and they respectively represent different structure of a image.Therefore,it’s very easy to extract the structure and detail information of the original image.We can say that the two-dimensional wavelet decomposition provides a good direction selectivity for image analysis.

The definition of wavelet transform is that performing displacement of τ to a function ψ （t） known as basic wavelet(also called mother wavelet),and then doing inner product in different scales of a with signal x（t） which is to be analysed.

In it,ψ*（ω） is the Fourier transformation form of ψ*（t）.

A.Haar Wavelet

There are many commonly used wavelet bases in wavelet transform,such as Daubuchies Wavelet,Mexican Hat Wavelet and Morlet Wavelet.These wavelet bases have good resolution and smoothness,but their common shortcoming is the large amount of calculation.So,in the algorithm proposed in this paper we choose the Haar wavelet,the reason is:

·It’s simple to use Haar wavelet implementation;

·It is fast in the speed of operation to carry out wavelet decomposition with Haar wavelet;

·With little memory overhead,wavelet transform can be completed in their own position in the Haar wavelet decomposition process.

B.Haar Wavelet Decomposition Algorithm

Using Haar wavelet decomposition,we can get the Integer representation of wavelet coefficients which is benefit to the recovery of the image without distortion.

As for the Haar wavelet decomposition,the coefficients for standardized wavelet function are as follows:

Equivalent representation in the frequency domain is as follows:

Using Haar wavelet to perform horizontal and vertical filtering on image,we can get the Haar wavelet decomposition of two dimensions.The two-dimension Haar wavelet decomposition can be reduced to four sub-band template operators:

The four operators can decompose a image into four subbands:LL,LH,HL and HH.LL reflects the low frequency information of level and vertical direction;LH reflects the low frequency information in horizontal direction and high frequency information in vertical direction;The HL reflects the high frequency information in horizontal direction and low frequency information in vertical direction;HH reflects the high frequency information of both vertical and vertical direction.The low frequency part is generally reflects the smooth area,but brim,details,noise generally exists in the high frequency part.Therefore,the processing of low frequency part won't make the details of a image fuzzy,and won't amplify the noise of original image.

Fig.1 Schematic diagram of two-dimensional wavelet decomposition

2 Algorithm Simulstion and Experimental Results

A.The Design of Algorithms

1）Firstly perform two dimensions multi-scale wavelet decomposition to short exposure image and long exposure image.

2）Then do summation and averaging processing for the lowfrequency coefficients of the two acquired images.

3）Take the maximum of the high frequency coefficients of these two images.

4）perform wavelet coefficients fusion to the acquired low-frequency coefficients in 2)and high frequency coefficients in 3).

5）Carry out wavelet inverse transform and wavelet reconstruction,then realize the enhancement of wide dynamic image.

B.The Experimental Results

This paper compares the processing results using the software of Photomatix with the processing results by the algorithm mentioned,and shows the images as follows:

Fig.2 Short exposure image

Fig.3 Long exposure image

Fig.4 Photomatix processing results

Fig.5 The algorithm enhanced image

The second group:

Fig.6 Short exposure image

Fig.7 Long exposure image

Fig.8 Photomatix processing results

Fig.9 The algorithm enhanced image

In this experiment,we take a short exposure image as shown in Fig.2 and a long exposure image in Fig.3.Firstly,use Photomatix software to conduct high dynamic image synthesis for the two pictures above,and get Fig.4.Then synthetize Fig.2 and Fig.3 by using the proposed algorithm to get a high dynamic range image,as shown in Fig.5.

Compare the results of Fig.4 and Fig.5 and so do to Fig.8 and Fig.9:The processed image of Photomatix is partial color seriously,and it has a great noise and severe distortion.However the processed image by the algorithm presented in this paper is more close to the scene,and can reflect the detail information of the image.

It can be seen from the graph that the detail characteristics of the high dynamic image are more evident,the over exposure part of long exposure image also have greatly improvement.

3 Conclusions

This paper reflects the superiority of this algorithm through comparing the processing results using Photomatix software with the results of the proposed algorithms.In this algorithm,It is the first time that wavelets transform method is used in high dynamic image fusion,and it widen the road for the development of the high dynamic fusion.At the same time the high dynamic development needs everyone’s positive effort and continue study.

The research and development of high dynamic range imaging technology provides a strong motivation to move forward for the development of digital images in the form of high-quality and large information volume.At present in our country,the development of HDR technology has not received enough attention.This paper is only as a preliminary study the authors carried out in this area,hoping to draw people’s attention.

Acknowledgment

The authors would like to acknowledge the Intelligence Video Department of Qingdao Hisense Transtech company for providing financial support for this work.

［1］De Neve,S.Goossens,and B.Hiep Luong Philips,An improved HDR image synthesis algorithm[C].ICIP,2009:1545－1548.

［2］Li Zhang,A.Deshpande,and Xin Chen,Denoising vs.deblurring:HDR imaging techniques using moving cameras[C].CVPR,2010:522－529.

［3］T.Poonnen,Li Liu Karia,and K.V.Joyner,A CMOS video sensor for High Dynamic Range(HDR)imaging Signals[C]//42nd Asilomar Conference on Systems and Computers,2008:853－856.

［4］Zhongkang and Lu Rahardja,Realistic HDR tone-mapping based on contrast perception matching[C].ICIP,2009:1789－1792.

［5］Sugiyama,N.Kaida,H.Xinwei,and Xue Jinno.HDR image compression using optimized tone mapping model[C].ICASSP,2009:1001－1004.

［6］Khan,I.R.Huang,and Z.Farbiz,HDR image tone mapping using histogram adjustment adapted to human visual system[C].ICICS,2009:1－5.

［7］Jinno,T.Okuda,and M.Motion,blur free HDR image acquisition using multiple exposures[C].ICIP,2008:528－531.

［8］Zhe Wendy,Wang Jiefu,and Zhai Tao Zhang,Interactive tone mapping for High Dynamic Range video[C].ICASSP,2010:1014－1017.

［9］Barakat,N.Hone,A.N.Darcie,and T.E,Minimal-Bracketing Sets for High Dynamic Range Image Capture[J].IEEE Transactions on Image Processing,2008:1864－1875.