Image Fusion Algorithm Based on Contrast Pyramid and its Performance Evaluation

Abstract

Contrast pyramid algorithm is put forward in this paper. The human visual system is sensitive to contrast information of image, so contrast pyramid algorithm would outstanding the contrast of image. The algorithm consists of creation process of Gauss Pyramid, the process of creating contrast Pyramid and reconstruction process of clear image. Simulation by MATLAB was completed in multi-focus image, multi-modality image and color image. Objective evaluation index such as mean, standard deviation, entropy and average gradient was calculated Simulation results and index show that the contrast pyramid algorithm has advantage of projecting the contrast of image, especially in color image fusion.

A multi-focus image fusion method via region mosaicking on Laplacian pyramids

“…In Fig 7(e) and 7(f) , RMLP is compared with six state of the art methods that include pulse coupled neural network (PCNN) [ 1 ], nonsubsampled contourlet transform (NSCT) [ 21 ], interactive digital photomontage (IDP) [ 32 ], Laplacian pyramid (LP) [ 11 ], the contrast pyramid (CP) [ 16 ], Discrete Wavelet Transform (DWT) [ 46 ], the Shift-invariant Discrete Wavelet Transform (SIDWT) [ 25 ], Principal Component Analysis(PCA) [ 24 ] and sparse reconstruction method [ 29 ]. PCNN is implemented according to the descriptions in [ 1 ].…”

“…The initial layer of Gaussian pyramid, G 0 , is the source image I ( x , y ). In [ 16 ], the k th layer of Gaussian pyramid can be expressed as: Where w ( m , n ) is a window function with Gaussian low-pass filter, K is the number of layers, R k and C k are the number of columns and rows in the k th layer pyramid respectively. The bottom pyramid G 0 constitute the Gaussian pyramid with other pyramids G 1 , …, G K -1 .…”

“…After being combined by certain guidelines, the fused coefficients are inversely transformed back to spatial domain to obtain the expected fusion image. Many transform domain based algorithms are brought forward, such as the Laplacian pyramid (LP) [ 11 ], the Discrete Wavelet Transform (DWT) [ 12 ] [ 13 ], the gradient pyramid [ 14 ] [ 15 ], the contrast pyramid (CP) [ 16 ], the ratio-of-low pass pyramid [ 17 ], the Shift-invariant DWT (SIDWT) [ 18 ], the Complex Wavelet Transform (CWT) [ 19 ] [ 20 ] and the Contourlet Transform [ 21 ] [ 22 ] [ 23 ].…”

“…When applying the block based methods, source images are firstly divided into N × M blocks. After comparing the sharpness of the blocks at corresponding position, the blocks with higher sharpness are spliced together to produce an all-in-focus image [ 16 ]. This kind of methods may result in block artifacts.…”