A New Strategy to Improve Image Fusion Effect

Abstract

The purpose of image fusion is to combine information from several different source images to one image, which becomes reliable and much easier to be comprehended by people. Based on analyzing the relations of average and standard deviation of the two or more source images, a new strategy to improve image fusion effect and a new evaluation measure named RAS (the ratio between average and standard deviation) are proposed in this paper. We apply wavelet transform to decompose an image into low-frequency sub-image and high-frequency sub-images and apply different fusion rules respectively to low-frequency sub-image and high-frequency sub-images. According to subjective evaluation and objective criteria, such as entropy, root mean square error (RMSE), peak-to-peak signal-to-noise ratio (PSNR), RAS, the proposed strategy is very effective and universal to some extent for fusing a class of images whose average and standard deviation are approximately equal respectively through extensive experiments.

A New Deep Learning Based Multi-Spectral Image Fusion Method

“…The IR and VIS images are captured by different imaging modalities, while the transform domain fusion method is suitable to produce less unexpected artifacts for good consistency with human visual perception. To cope with this issue, we decomposed both IR and VIS images by using a 3-level 2-D Haar wavelet transform [ 23 ], and the input image pairs were decomposed into low and high-frequency sub-bands. Since the size of the original image is down sampled during wavelet transform in each level, the weight map is scaled to match the size of down sampled images.…”

“…The number of levels is selected by considering these factors. Details of wavelet transform-based image decomposition and reconstruction are introduced in [ 23 ]. The wavelet transform-based fusion scheme is illustrated in Figure 7 .…”

Adaptive Image Fusion Scheme Based on Contourlet Transform, Kernel PCA and Support Vector Machine

“…The drawback of this technique is that the resulting image will have the reduced contrast and possibly contain noise. Other common methods include Laplacian and Gaussian Pyramid [5,7], ratio of low pass pyramid [6], and the discrete wavelet transform (DWT) [1,2,14]. The idea behind the above mentioned techniques is to decompose the source images by applying the transformation and then combining all the decompositions based on some fusion rules to form a single composite representation, then fused image can be obtained by applying the inverse transform.…”

“…pixel level, feature level and decision level [4]. In Pixel level fusion, the pixels in fused images are computed from pixels in the source images [14]. In Feature level fusion, first feature extraction step is performed on the source images, followed by the fusion step that is based on some selection criteria.…”