Multifocus Image Fusion Using Wavelet-Domain-Based Deep CNN

Abstract

Multifocus image fusion is the merging of images of the same scene and having multiple different foci into one all-focus image. Most existing fusion algorithms extract high-frequency information by designing local filters and then adopt different fusion rules to obtain the fused images. In this paper, a wavelet is used for multiscale decomposition of the source and fusion images to obtain high-frequency and low-frequency images. To obtain clearer and complete fusion images, this paper uses a deep convolutional neural network to learn the direct mapping between the high-frequency and low-frequency images of the source and fusion images. In this paper, high-frequency and low-frequency images are used to train two convolutional networks to encode the high-frequency and low-frequency images of the source and fusion images. The experimental results show that the method proposed in this paper can obtain a satisfactory fusion image, which is superior to that obtained by some advanced image fusion algorithms in terms of both visual and objective evaluations.

Spatiotemporal Fusion of Land Surface Temperature Based on a Convolutional Neural Network

“…CNN-based fusion methods utilize end-to-end CNN to automatically form the nonlinear relationship between observed pairs of coarse and fine spatial resolution image and use this to predict the target fine spatial resolution image. Because high-level features (e.g., edges of objects and their mutual relations) [37] which contain abundant semantic information can be extracted from the input data through CNN, the established relations between input and output are more practical, thus more favorable results can be acquired. But there are also some limitations when they are adopted for the generation of LST products.…”

Fast Multi-Focus Fusion Based on Deep Learning for Early-Stage Embryo Image Enhancement

“…Commonly used techniques are wavelet, curvelet and contourlet transforms, neighbour distance, Laplacian pyramid or gradient pyramid [ 13 , 14 ]. Deep learning methods (specifically CNN based approaches) are often incorporated to solve blurring-effect problems through the ability to learn the focus measure to recognize the focused and defocused pixels or regions in source images [ 15 , 16 ]; to learn the fusion operation to fuse a pair without the need for ground truth fused images [ 17 , 18 ]; to learn the direct mapping between the high-frequency and low-frequency images of the source and fusion images [ 19 ], and so forth. All these methods can be used to obtain the best focus image from a set of captured microscopic images, but the performance time is the essential factor for the embryo selection task.…”

Image Super Resolution Using Fractal Coding and Residual Network

“…(3) for i � 1 to N r do (4) Take an "uncoded" R i from C, size is N × N, exhaustively find the optimal domain block D i in all image blocks of size 2N × 2N and make it the smallest patch error e under the approximation of equation 5; (5) if e ≤ ϑ and N × N ≤ N stop × N stop do (6) Record the size of the range block and the fractal code W; (7) Mark as "encoded"; (8) else (9) Decompose R i into four smaller sub-blocks R a , R b , R c , and R d , all marked as "uncoded" and added to C; (10) Complexity methods: SRCNN [16], VDSR [17], EDSR [18], RDN [31], and PASSR [44].…”

“…However, these methods have limited ability to extract and express features during the learning process. In recent years, the method of deep learning is applied to many fields [16,17], and many researchers have introduced the convolutional neural network (CNN) into SR reconstruction and learning end-to-end mapping between LR and HR images by relying on external data sets; for example, SRCNN [18], VDSR [19], and EDSR [20] show superiority. In particular, the proposed residual network makes the data transmission between the networks smoother, which makes the depth of the network increase and the reconstruction effect better.…”