Abstract:
Image fusion has its applications in many fields such as computer vision,
automatic object detection, robotics, remote sensing, military and law enforcement,
medical imaging and manufacturing. The objective of image fusion is to generate a
resultant fused image from a set of input images (of the same scene) which describes the
scene better than any single input image with respect to some relevant properties. The
fused image is obtained by extracting all the useful information from the source images
while not introducing artifacts or inconsistencies which will distract human observers or
the following processing. For this purpose a new image fusion technique that is actually
integration of multi-scale wavelet transform, gradient and mathematical morphology
schemes, has been proposed. The proposed scheme’s implementation mainly consists of
five steps. The first step is the application of discrete wavelet transform on the set of
multifocused source images. The second step deals with the computation of local gradient
of each detailed wavelet coefficient block. Finding of image activity level is the next step.
Generation of binary decision map takes place based on image activity levels obtained at
the previous step. Different morphological operations have been performed on binary
decision map that separate the focus and defocused parts of the input images. Finally,
fused image has been achieved by using the processed binary decision map.
The empirical results on standard test images (i.e. Lena, Barbara, Gold Hill and
Peppers) provide higher Peak Signal to Noise Ratio (PSNR) and smaller Root Mean
Square Error (RMSE) values than some of the previous approaches. These fusion results
strengthen the idea of using combination of multi-scale wavelet transform, gradient and
mathematical morphology schemes for multifocus image fusion.
MATLAB 7.0 has been used for the implementation of the proposed approach.
Experiments have been carried out on a variety of standard greyscale images with
different defocus parts.
