25-10-2016, 03:11 PM
1461136136-1407.3986.pdf (Size: 504.28 KB / Downloads: 15)
Abstract- Image Fusion is the process in which core
information from a set of component images is merged to
form a single image, which is more informative and
complete than the component input images in quality and
appearance. This paper presents a fast and effective image
fusion method for creating high quality fused images by
merging component images. In the proposed method, the
input image is broken down to a two-scale image
representation with a base layer having large scale
variations in intensity, and a detail layer containing small
scale details. Here fusion of the base and detail layers is
implemented by means of a Local Edge preserving
filtering based technique. The proposed method is an
efficient image fusion technique in which the noise
component is very low and quality of the resultant image
is high so that it can be used for applications like medical
image processing, requiring very accurate edge preserved
images. Performance is tested by calculating PSNR and
SSIM of images. The benefit of the proposed method is
that it removes noise without altering the underlying
structures of the image. This paper also presents an image
zooming technique using bilinear interpolation in which a
portion of the input image is cropped and bilinear
interpolation is applied. Experimental results showed that
the when PSNR value is calculated, the noise ratio is found
to be very low for the resultant image portion.
INTRODUCTION
Image fusion is an efficient method for
creating high quality images in a wide range of
applications involving remote sensing and medical
image processing. In image fusion, component images
of the same scene can be merged into a single fused
image. More elaborate information about the visual is
obtained which is useful for human and machine
perception. The requirements of a good image fusion
method are the following. First, it should preserve most
of the useful information of component images
including the edges. Second, it would not produce
image artifacts. Third, it should produce fast output and
could be able to tolerate to conditions like noise in the
input. A major drawback of guided filtering based
image fusion is that it may over-smooth the resulting weights, which is not good for image fusion. To solve
the problem mentioned above, an image fusion method
with LEP (Local Edge Preserving) is proposed here.
From experimental results it is found that the proposed
LEP filtering based image fusion method gives much
better performance when compared with other
approaches. There are many advantages of the
proposed image fusion method .It is a fast two-scale
fusion method which does not rely heavily on a
specific image decomposition method .Pixel saliency
and spatial context for image fusion are achieved by a
novel weight construction method. In this method the
balance of pixel saliency and spatial consistency are
obtained by adjusting the parameters of the LEP filter.
Bilinear interpolation is used to improve the
quality of the image generated by using LEP filtering
while zooming the image. The capability of
interpolation to estimate the values of a continuous
function from discrete samples is utilized here. This is
implemented by determining the grey level value from
the weighted average of four nearest pixels to the input
coordinates, and assigning this value to the output
coordinate
Experimental results showed that when image
fusion with LEP filtering and bilinear interpolation is
used, the parameters such as sharpness, naturalness,
SSIM and PSNR values are preserved and the noise
ratio is very low.
The existing image fusion technique based on
guided filtering has a lot of drawbacks. When guided
filtering is used for image fusion, the edges of resultant
image are not well preserved, or in other words the
pixels representing the edges are highly distorted.
Second major drawback of existing fusion method is
that it takes more time to produce image output. This is
because of the time complexity associated with the
guided filtering algorithm. While processing high
resolution images, the existing system yields poor
output. There is significant loss of quality of the
original image. Third significant problem is the
existence of artifacts in the fused image.
CONCLUSION
In this paper we have proposed an image
fusion method based on Local Edge Preserving filter.
As the name indicates this algorithm can efficiently
preserve the pixels representing the edges of resultant
image. Most of the pixel information representing the
edges of the image are retained and there is little loss of
information. This method utilizes the average filter to
get the two-scale representations, which is simple and
more effective. More importantly, weight optimization
is achieved by using the LEP filter in such a way as to
make full use of the correspondence between adjacent
pixels. Experiments show that when the proposed image
fusion method is applied, the information contained in
the component images is well preserved. The
processing of medium to high resolution images using
LEP filtering yields good output. The quality of the
resultant image is also high in this method and the
presence of artifacts is very rare. The application of
Bilinear Interpolation technique further improves the
quality of fused image when zooming is applied. The
proposed method applies Image Zooming with Bilinear
interpolation in addition to LEP filtering. When Bilinear interpolation is applied, the parameters such as
sharpness, naturalness, SSIM and PSNR values are
preserved and the noise ratio is found to be very low.
The LEP filtering based image fusion with Bilinear
interpolation is good for applications requiring high
quality images like medical image processing.
Experiments show that the proposed method using LEP
filtering is many times faster than the existing guided
filtering based image fusion. The overall performance
of the proposed method can be improved in future
researches by adaptively choosing the parameters of the
LEP filter thereby obtaining supreme quality images
even after zooming several times.