Anda di halaman 1dari 5

International Journal of Computer Technology and Electronics Engineering (IJCTEE) Volume 1, Issue 1

Implementation of Hybrid Image Fusion Technique Using Wavelet Based Fusion Rules
Paresh Rawat, Sapna Gangrade, Pankaj Vyas

Abstract Here we focused on Multi Resolution fusion by adopting discrete wavelet transform (DWT) technique. This paper highlights wavelet based pixel-level fusion. In this paper we adopt a novel approach to decompose the original images into high and low frequency parts to the smallest pixel and then fuse both the parts separately using same fusion rules to get an accurate, high resolution image with preserved spectral characteristics.In this paper work has been done in three stages.In first stage we have compared various wavelet filters by using a standard fusion technique In second stage a novel image fusion method known as Hybrid image fusion is implemented. In third stage we have implemented proposed algorithm on stimulated images. The experimental set up demonstrates that the proposed approach performs better in both subjective and objective qualities. Index Terms Multiresolution, Image Fusion, Hybrid Fusion, DWT, pixel level fusion.

There are three methods in multi resolution fusion scheme, namely Pixel based fusion, Area based fusion and Region based fusion methods. Many methods exist to perform image fusion. The very basic one is the high pass filtering technique. Later techniques are based on DWT, uniform rational filter bank, and Laplacian pyramid. B. Objectives of research work In this paper we perform the following task: (1) Various wavelet filters are used to preserve the edges and details of the images as well as to efficiently reduce noise in multi focus images. This research work proposes comparison of various wavelet filters to achieve optimal wavelet filter on the basis of minimum Mean Square Error value and maximum Signal to Noise Ratio value (2) Results of various fusion techniques are extracted and then these results are again fused by implementing Hybrid Image Fusion method. (3) Proposed methodology of Hybrid Image Fusion,which has minimized errors and also has presented better quality results as compare to conventional image fusion method of Diagonal Averaging. C. Platform The MATLAB package offered a significant advantage as it has provided access to the Image Processing Toolbox that offers a comprehensive suite of standard algorithms and graphical tools for image processing, analysis and visualization . However, using the Image Processing Toolbox in MATLAB would enable instant access to standard image processing algorithms and would also allow the application to be easily ported to any environment that is supported by MATLAB, so MATLAB was chosen as the computational package that would be used to develop the application. Procedure for Paper Submission II. BACK GROUND KNOWLEDGE A. Discrete Wavelet Transform The discrete wavelet transform (DWT) of image signal produces image representations which provides better spatial and spectral localization of image formation compared with other multi scale representations such as Gaussian and Laplacian pyramid[4] .

I. INTRODUCTION The fast development of digital image processing leads to the growth of feature extraction of images which leads to the development of Image Fusion. A. Image Fusion Image Fusion is defined as the process of combining two or more different images into a new single image. In this process the new single image so formed retain important features from each image with extended information content. Image fusion is a powerful tool which is used to enhance the quality of image. Image fusion technique is classified into two categories Direct Image Fusion and Multi resolution Image Fusion. Here, it is focused on Multi resolution Image Fusion techniques based on pixel level fusion methods. Multi Resolution fusion uses pyramid or wavelet transform at multi scale for the representation of the source images [9].

Manuscript received july 12, 2011. Paresh Rawat,, Assistant Professor,Department of Electronics and Communication,TRUBA I.E.I.T.,Bhopal M.P.India. (e-mail: parrawat@yahoo.com). Sapna Gangrade, Assistant Professor,Department of Electronics and Instrumentation ,TRUBA I.E.I.T.,Bhopal M.P.India.. Pankaj kumar Vyas, Assistant Professor,Department of Electronics and Communication,TRUBA I.E.I.T.,Bhopal M.P.India. (e-mail: pankajv10@gmail.com).

International Journal of Computer Technology and Electronics Engineering (IJCTEE) Volume 1, Issue 1 Recently, DWT has attracted more and more interest in image fusion. An image can be decomposed into a sequence of different spatial resolution images using DWT[5]. C. Pixel Level Fusion This section focuses on the so-called pixel level fusion process, where a composite image has to be built of several input images[1]. In pixel-level image fusion, some generic requirements can be imposed on the fusion result: a) The fusion process should preserve all relevant information of the input imagery in the composite image (pattern conservation) b) The fusion scheme should not introduce any artifacts or inconsistencies which would distract the human observer or following processing stages. C.1 Pixel Based Fusion Rules Figure1: The pyramidal decomposition shown up to two levels B. Wavelet Based Image Fusion Image fusion is the process of combining two or more source images into composite images with extended information content[3]. It is an intensity biased image fusion technique. In which we directly work on the intensities of individual pixels [1]. Let A (x, y) and B (x, y) are images to be fused, the decomposed low frequency sub images of A (x, y) and B (x, y) are lAJ (x, y) and lBJ (x, y). Decomposed high frequency sub images of A (x,y) and B(x,y) are hAj,k(x, y) and hBj,k(x, y). (j and k is the parameter of resolution, where j=1,2,3.J for every j and k=1,2,3 k for every k) There are different pixel level fusions methods like as: 1)Pixel Averaging All the four sub bands of the fused image F is simply acquired by averaging the wavelet coefficients of source images A & B. Figure 2 Image fusion Fj,k = (Aj k +Bj,k )/2 & FJ = (lAj+ lBj) /2 Image fusion technique [2]is a powerful tool for extracting high quality information from large number of remotely sensed images and limiting redundancy among these images. Image fusion technique, with an image fusion engine to organize, join, and combine multi source and multi temporal data, provides a powerful tool for these data processing problems[9]. Wavelet transform is first performed on source images. Then a fusion decision map is generated based on a set of fusion rules. Then fused wavelet coefficient map can be constructed from the wavelet coefficients of the source images according to the fusion decision map. Finally the fused image is obtained by performing the inverse wavelet transform[2]. The DWT based techniques became popular due to their multi-resolution properties[5]. The main issue in wavelet fusion technique is the selection of decomposition level and the optimum wave filter. (i) 2)Pixel Level Maximum (PLM) All the four sub bands of the fused image F is simply formed by taking the wavelet coefficients from source images which is having the maximum value. Fj,k = max (Aj,k Bj,k ) & FJ = max (lAj, lBj) (ii)

LL

HL

L H L H

L H LH HH LH

H L H H

HL

HH

3)Pixel absolute values (Method 3) Since larger absolute transform coefficients correspond to sharper brightness changes, the good integration rule is to select, at every point in the transform domain, the coefficients whose absolute values are higher. Fj,k = max (abs (Aj,k), abs (Bj,k )) & FJ = max (abs (lAJ), abs (lBJ)

(iii)

4)Pixel averaging / absolute This method uses averaging to fuse low frequency sub bands and selects the absolute maximum value for high frequency bands. Fj,k = max(abs (Aj,k ), abs(Bj,k )) , & FJ = (LAj+LBj )/2 5)Pixel diagonal averaging / absolute It takes the average of approximation and diagonal details of source image to form the LL and HH sub band and uses absolute maximum criterion to form the LH and HL sub bands of fused image.

Figure. 3 Wavelet Based Image Fusion

International Journal of Computer Technology and Electronics Engineering (IJCTEE) Volume 1, Issue 1 Finally these two results are fused together hence produces III. HYBRID IMAGE FUSION the final hybrid image. Hence the result so obtained not only The proposed Hybrid fusion method integrates various pixel level fusion rules in a single fused image[12]. Pixel have better quality of image as compare to the traditional based rules operate on individual pixels in the image, but does methods, but also removes the drawbacks of traditional not take into account some important details like edges, methods. boundaries and salient features larger than a single pixel. Pixel level rules may reduce the contrast in some images and does not always succeed in effectively removing ringing artefacts and noise in source images. The inadequacies of these types of fusion rules point to the importance of developing a Hybrid algorithm to improve the visual quality by combining the advantages of four pixel based methods. The proposed Hybrid Image Fusion method is based on Pixel level image Fusion methods. These methods are basically based on various standard Pixel level fusion rules [1]. Combination of various fusion rules is done to get better quality final fused image. Here the image fusion techniques used are based on wavelet transformation. First level and second level decomposition of original image is based on Discrete Wavelet Transformation (DWT)[5]. In this hybrid method, first wavelet decomposition of the input source images is performed up to level N. The low pass and high pass sub-bands are then fused using different pixel level fusion methods. Then the inverse wavelet transformation is performed to get full size fused images [4]. The results of pixel level minima and pixel level averaging are fused together. The results of pixel level maxima and diagonal averaging are fused together. And finally these two results are fused to get the final fused result. The results of Pixel level minima and Pixel level averaging are fused together. The results of Pixel level maxima and Diagonal averaging are fused together. IV. SIMULATION RESULTS In this section some of experimental results of our work on wavelet based Image Fusion are discussed. Input image database is taken in different environment through a digital camera, the multi focused environment is generated using different light effects. In the proposed hybrid method, first wavelet decomposition of the input source images is performed up to level second level using discrete wavelet transform. A. Result of Wavelet Filters In the first stage the results of the different wavelet filters for an simplest and specific fusion technique PLM are presented. The performance of the wavelet filters are compared using mean square error and Signal to noise ratio
Fused 1 img Fused 1 img

(a)

Fused 1 img

(b)

Fused 1 img

(c)

(d)

(e) (f) Figure 5 Results of wavelet filters for Drawing room image a) Original image1; b) Original image2 c) dB4 fused image; d) Haar fused image, e) Coiflet fused image; f) Symlet fused image
Fused 1 img Fused 1 img

(a)

(b)

Fused 1 img

(c)
Fused 1 img

(d)

(e) (f) Figure 6 Results of wavelet filters for Showpiece image (a) Original image1;(b) Original image2 (c) dB4 fused image; (d) Haar fused image(e) Coiflet fused image ; (f) Symlet fused image Table 4.1 MSE for various filters
IMAGES DRAWING ROOM SHOWPIECE

Figure 4 Hybrid Image Fusion

Wavelet Filters db4 Haar Coiflets Symlet

30.325 4.708 11.072 15.091

31.128 4.567 12.228 16.883

International Journal of Computer Technology and Electronics Engineering (IJCTEE) Volume 1, Issue 1 This can be seen that Table 4.1 and Table 4.2 shows the values of MSE and SNR respectively. The wavelet decomposition of the input source images is performed up to level N. The Low pass and high pass sub-bands are then fused using different pixel level fusion. Table 4.2 SNR for various filters Normal Pixel based method Minima Maxima Averaging (F1) (F2) (F3) Hybrid Pixel based method F5 = (F1) F6 = (F2) Final Fused Fused (F5) (F3) (F4) Fused (F6) 18.511 14.019 12.247 5.2348 5.0923 4.9297 26.835 27.27 26.481 22.236 19.468 17.028 6.0972 5.3028 4.9272 15.853 14.73 14.196 21.01 15.981 14.853 8.1378 6.0518 8.8248 7.2651 7.2616 7.1042 8.9542 9.3324 10.497 7.2833 9.1591 9.1967 5.0663 4.5648 4.6001 4.7563 4.7435 4.6664 6.41 5.9155 5.9644 13.193 9.7343 11.482 6.4884 6.324 6.128 15.183 15.601 16.731 12.79 14.367 13.676 6.1229 5.3792 5.2037 7.6569 7.5349 7.2039 10.19 8.989 8.9161 methods. Then the inverse wavelet transformation is performed to get full size fused images. The results of pixel level minima and pixel level averaging are fused together. The results of pixel level maxima and diagonal averaging are fused together.

S. No .

Images

Diagonal Averaging (F4) 4.5273 3.2809 6.2027 4.1186 4.3317 4.416 5.0407 5.3164 6.3676 3.8049 5.0029 5.2216 3.5933 3.4459 3.7935 3.983 4.247 4.9167 5.276 5.387 5.7407

128x128 Clock1 1 2 clock 2 256x256 512x512 128x128 256x256 512x512 128x128 3 clock 3 256x256 512x512 128x128 4 draying room 256x256 512x512 128x128 5 Croud 256x256 512x512 128x128 6 7 Home Statue 256x256 512x512 128x128 256x256 512x512

16.134 11.648 9.8593 3.9755 3.8724 3.74 30.094 31.286 29.505 21.717 17.69 15.414 4.9274 4.2111 3.8501 29.516 22.613 21.176 50.006 21.021 17.009

14.673 12.379 9.9267 12.187 10.527 9.453 15.407 15.396 14.655 19.871 17.985 15.346 5.6473 4.8169 4.5731 4.6874 4.5206 4.191 6.3796 5.5955 5.5183

15.831 13.347 12.812 7.1748 6.9213 6.8013 17.287 17.314 17.133 16.117 15.499 13.638 6.2967 5.8252 5.7889 8.5683 8.2978 8.0614 11.002 9.4763 9.441

Original image 2 Diagonal fused Hybrid Fused Figure 7. Result of fusion for size of 12x128 for clock 1 image error and signal to noise ratio after resizing the original And finally these two results are fused together hence images are compared. produces the hybrid image. The images are resized to check the performance of the fusion technique. The Mean square

wavelet Original image 1

International Journal of Computer Technology and Electronics Engineering (IJCTEE) Volume 1, Issue 1 [10] Milad Ghantous, Soumik Ghosh and Magdy Bayoumi, A V. CONCLUSION AND FUTURE WORK In the current research work, in first stage we have firstly compared various filters. It is found that among various filters better results are extracted at lower orders of filters and that too Haar filter showed the best results with minimum MSE and maximum SNR. Secondly, as in this research work is being done on Pixel Level Fusion rules such as Pixel Level Maxima, Pixel Level Minima, Pixel Level Averaging and Pixel Level Diagonal Averaging. As the results of Diagonal Averaging method show better features comparatively, but with maximum MSE and minimum SNR value. Hence our target was to achieve better features of an image with minimum MSE and maximum SNR. To achieve the desired target an algorithm is proposed in the second stage of the current work. In the third stage various stimulated images such as football.jpg, and peppers.png, images are taken. The results showed that proposed Hybrid fusion method has minimized MSE and has maximized SNR with respect to Diagonal averaging method for stimulated images also. The results obtained from proposed hybrid fusion algorithm can further be applied to DT-CWT method to segment the features of the input images, either jointly or separately, to produce a region map. REFERENCES
[1] Yang bo., Zing Zhong-liang and Zhao Hai, Review of Pixel-Level Image Fusion, J. Shanghai Jiaotong Univ. (Sci.), 2010, 15(1): pp.6-12 K. Kannan, S. Arumuga Perumal, K. Arulmozhi, Area level fusion of Multi-focused Images using Multi-Stationary Wavelet Packet Transform, Int. Journal of Computer Applications Volume 2 No.1, May 2010. pp. 314-318 Tanish Zaveri, Mukesh Zaveri, A Novel Two Step Region Based Multifocus Image Fusion Method , International Journal of Computer and Electrical Engineering, Vol. 2, No. 1, February, 2010,1793-8163. pp. 86-91[9] Hong Zheng, Dequan Zheng, Yanxiang HU, Sheng LI, Study on the Optimal Parameters of Image Fusion Based on Wavelet Transform, Hindawi Publishing Corporation, EURASIP Journal on Advances in Signal Processing,Article ID 579341, doi:10.1155/2010/579341. System6:1(2010),pp. 131-137 Anjali Malviya, S.G. Bhirud, Image Fusion of Digital Images, Int. Journal of Recent Trends in Engineering, Vol 2,No 3, Nov. 2009.pp. 146-148 Rohini S. Asamwar, Kishor M. Bhurchandi, Abhay S. Gandhi, Interpolation of Images Using Discrete Wavelet Transform to Simulate Image Resizing as in Human Vision, International Journal of Automation and Computing 7(1), February 2010, DOI: 10.1007/s11633-010-0009-7. pp. 9-16 M.A. Mohamed and B.M. El-Den, Implementation of Image Fusion Tec hniques Using FPGA, IJCSNS International Journal of Computer Science and Network Security, Vol.10 No.5, May 2010, 95. A.K.M. Fazlul Haque, Md. Hanif Ali, and M.Adnan Kiber, Biomedical Image Analysis Using Wavelet Tools For Emergency Medical Applications, ARPN Journal of Engineering and Applied Sciences,Vol.5, No.1, January 2010,ISSN 1819-6608, pp15-20 Hong Zheng, Dequan Zheng, Yanxiang HU, Sheng LI, Study on the Optimal Parameters of Image Fusion Based on Wavelet Transform, Hindawi Publishing Corporation, EURASIP Journal on Advances in Signal Processing,Article ID 579341, doi:10.1155/2010/579341. System6:1(2010),pp. 131-137. Susmitha Vekkot and Pancham Shukla , A Novel Architecture for Wavelet based Image Fusion Journal of World Academy of Science, Engineering and Technology, 2009. pp. 32-33. Gradient-Based Hybrid Image Fusion Scheme Using Object Extraction, 978-1-4244-1764-3/08/$25.00 2008 IEEE ,ICIP 2008. pp. 1300-1303. Jun Kong, Kaiyuan Zheng, Jingbo Zhang, Xue Feng, Multi-focus Image Fusion Using Spatial Frequency and Genetic Algorithm, IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.2, February 2008. pp. 220-224.

[11]

[2]

[3]

[4]

[4]

[5]

[6]

[7]

[8]

[9]

Anda mungkin juga menyukai