CN105225213A - A kind of Color Image Fusion method based on S-PCNN and laplacian pyramid - Google Patents

Abstract

The invention provides a kind of Color Image Fusion method based on S-PCNN and laplacian pyramid, proposes a kind of effectively Color Image Fusion based on simplification pulse-couple network (S-PCNN) and Lars, Lapland pyramid algorith.In HSV color space, utilize S-PCNN to add after local entropy carries out characteristic area cluster to H component, merge based on the pulse oscillating H component that figure realizes each source images frequently; Utilize laplacian pyramid to carry out resolution decomposition S, V component, then utilize different fusion rule to merge S, V component.Finally color space inverse transformation is carried out to new H, S, V component, achieve the fusion of RGB color image.Experimental result of the present invention shows, no matter algorithm of the present invention is in subjective vision effect, or objective evaluation standard is all better than other conventional Image Fusion.

Description

A kind of Color Image Fusion method based on S-PCNN and laplacian pyramid

Technical field

The invention belongs to technical field of image processing, be specifically related to a kind of Color Image Fusion method based on S-PCNN and laplacian pyramid.

Background technology

Image co-registration is the consistent branch of information fusion, is one of study hotspot of current information fusion.Color Image Fusion makes the image after fusion more accurate, comprehensive to the description of Same Scene, reliable.The research of current Color Image Fusion is relatively less, and human vision to color information can identification far above gray level image.Along with sensor technology improvement and provide, Color Image Fusion can be subject to increasing attention.Different brightness and different color combine expression image information by coloured image.So the fusion based on color space is generally merge respectively each component.But algorithms most in use such as IHS, weighting and PCA converter technique scheduling algorithm easily realize poor effect.And based on the blending algorithm of multiresolution analysis, generally first image conversion is carried out to multi-source image to be fused, then the coefficient after conversion is reconfigured.Difference according to picture breakdown mode is broadly divided into the blending algorithm based on pyramid transform, the blending algorithm based on wavelet transformation and the blending algorithm based on multiple dimensioned geometric transformation, and this class algorithm is used for the image co-registration of Pixel-level.Pixel-level merges the bottommost one deck being in image co-registration classification, and its each pixel is that other source images respective pixel determined.Pulse Coupled Neural Network (PCNN) to solve etc. the premium properties in field in image procossing, pattern-recognition, route according to it, be described as the 3rd ground artificial neural network.PCNN is a kind of new neural network being different from traditional artificial neural network, it is the feedback-type network be interconnected together by several neurons, formation PCNN neuron is a comprehensive dynamic non linear system, has the incomparable advantage of traditional artificial neural network than it.Laplacian pyramid algorithm is the image processing method that a kind of multiple dimensioned, multiresolution, multilayer are decomposed, it can by the key character of image (as edge, texture etc.) according to different Scale Decompositions on different decomposition layers, compared with simple image blending algorithm, it can obtain better syncretizing effect, has been widely used in the middle of image co-registration.

Summary of the invention

In order to solve the problems referred to above that prior art exists, the invention provides a kind of Color Image Fusion method based on S-PCNN and laplacian pyramid.

The technical solution adopted in the present invention is:

Color Image Fusion based on S-PCNN and laplacian pyramid of the present invention is studied, and comprises the following step:

A. first, by RGB color image space transforming good for registration to HSV color space;

B., after utilizing S-PCNN to carry out characteristic area cluster to H component, merge with the H component realizing each source images based on pulse oscillating frequency figure; H component processing procedure: first H component is sent into S-PCNN model and carry out iteration, obtain igniting and frequently scheme OFG, S-PCNN igniting corresponding to H component is more frequently schemed OFG and is carried out local entropy computing, obtain the eigenmatrix of OFG local entropy matrix as H component, finally contrast the size of the H component respective pixel local entropy LE of different source images, get the pixel with larger local entropy as fusion pixel;

C. image Laplacian pyramid, wherein, S, V component, utilizes laplacian pyramid to carry out resolution decomposition S, V component, then utilizes different fusion rule to merge S, V component;

D. last the inverse transformation of HSV color space is carried out to new H, S, V component that steps A, B, C obtain, finally obtain the RGB color image after merging.

Beneficial effect of the present invention is: the present invention is based on simplification pulse-couple network (S-PCNN) and propose a kind of effective Color Image Fusion with laplacian pyramid algorithm; First by RGB color image space transforming to HSV color space; In HSV color space, after utilizing S-PCNN+ local entropy to carry out characteristic area cluster to H component, the H component realizing each source images based on pulse oscillating frequency figure merges; Utilize laplacian pyramid to carry out resolution decomposition S, V component, then utilize different fusion rule to merge S, V component.Finally the inverse transformation of HSV color space is carried out to new H, S, V component, obtain the RGB color image after merging.The present invention is by some objective indicators, contrast with other several algorithms most in use, other conventional Image Fusion are all better than from subjective vision effect or objective evaluation standard, experimental result all shows that algorithm of the present invention can well preserve details and the color information of coloured image, and the coloured image after fusion is more clear, reliable.

Provided by the invention based on S-PCNN and laplacian pyramid algorithm, in conjunction with the hsv color model being suitable for human eye vision, propose a kind of effective Color Image Fusion.RGB color image is first converted to the HSV color space meeting human-eye visual characteristic by this algorithm; Use S-PCNN and local entropy to carry out cluster to H component characterization, then to it to merging, know that the PCNN model of mechanism to have the characteristic of provincial characteristics cluster to image because it has mammal visual cortex visual sense; Laplacian pyramid is carried out to S, V component, merged by different strategies, because it is effective tool characteristics of image being carried out to multiscale analysis, the band that its Pyramid transform data embody image leads to turriform filtering, and its size is consistent under each layer decomposition scale; Finally carrying out inverse transformation to merging the HSV chrominance component obtained, obtaining the RGB color image merged.Therefore, the present invention can utilize the advantage of Laplacian pyramid and S-PCNN, effectively merges RGB image.Experiment proves that the color fusion algorithms that the present invention proposes can merge not confocal coloured image preferably, and well can retain the details of source images, texture and principal character information.Illustrate that no matter algorithm of the present invention is at visual effect or objectively, is above better than other algorithms.

Accompanying drawing explanation

Fig. 1. be Color Image Fusion process flow diagram of the present invention;

Fig. 2 is embodiment of the present invention source Fig. 1;

Fig. 3 is embodiment of the present invention source Fig. 2;

Fig. 4 is fusion figure of the present invention.

Embodiment

As shown in Figure 1, the invention provides a kind of Color Image Fusion method based on S-PCNN and laplacian pyramid.Below in conjunction with drawings and Examples, the present invention is further detailed explanation.

Embodiment: a kind of Color Image Fusion method based on S-PCNN and laplacian pyramid, specifically comprises following steps:

The Color Image Fusion of A, S-PCNN and laplacian pyramid; First, RGB color image space transforming is HSV image, obtains H, S, V tri-components, and H, S, V tri-components are respectively: hue, saturation, intensity.H angle is measured, and span is 0 ° ~ 360 °; S representative color purity level, V represents image light levels, is the measurement to gray scale, and their scope is all 0 ~ l.RGB color image space transforming is as follows to the transformation relation formula of HSV color space:

$S=\frac{\max \left(R,G,B\right)-\min \left(R,G,B\right)}{\max \left(R,G,B\right)}---\left(2\right)$

V＝max(R，G，B)(3)

R in above formula is the red component in common rgb format image, G is green component, B blue component, above-mentioned formulae express be how RGB is transformed into HSV color space, calculated the representation of image in HSV territory, RGB and HSV is the different representations of same sub-picture.Because process H, S, V component is more effective than R, G, B component, so what adopt herein is to the process of H, S, V component, so, the image of rgb format be converted to the image of HSV form, then HSV tri-components be processed.

B, in HSV color space, after utilizing S-PCNN to carry out characteristic area cluster to H component, to merge with the H component based on each source images of pulse oscillating frequently figure experiment.H component processing procedure: first H component is sent into S-PCNN model and carry out iteration, obtain OFG (igniting is figure frequently), S-PCNN igniting corresponding to H component is more frequently schemed (OFG) and is carried out local entropy computing, obtains the eigenmatrix of OFG local entropy matrix as H component.Finally contrast the size of H component respective pixel local entropy (LE) of different source images, get the pixel with larger local entropy as fusion pixel.Wherein S-PCNN model is as follows;

F _ij(n)＝S _ij(4)

L _ij(n)＝V ^LΣ _klW _kjY _ijkl(n-1)(5)

U _ij(n)＝F _ij(n)[1+βL _ij(n)](6)

${\mathrm{\θ}}_{ij}\left(n\right)=e^{-{\mathrm{\α}}^{\mathrm{\θ}}}{\mathrm{\θ}}_{ij}(n-1)+V_{ij}^{\mathrm{\θ}}{Y}_{ij}\left(n-1\right)---\left(7\right)$

$Y_{ij}\left(n\right)=step\left(U_{ij}\left(n\right)-{\mathrm{\θ}}_{ij}\left(n\right)\right)=\left\{\begin{array}{c}1,U_{ij}\left(n\right)>{\mathrm{\θ}}_{ij}\left(n\right)\\ 0,otherwise\end{array}\right\}---\left(8\right)$

For neuron N _ij, the L passage that the input of F passage that formula (7) describes and formula (8) describe constitutes its acceptance domain, and wherein neuronic F passage accepts external drive and inputs S _ij, i.e. the pixel value of image, and L passage accepts neighborhood neuron N _klpulse excitation input Y _ijkl, W _kjfor neighborhood link power, and V ^lfor passage amplitude.Then in modulation domain, neuronic F passage exports and defines neuronic internal state value U, β with the output of L passage through the nonlinear modulation that is multiplied is the link strength that in modulation domain, L passage exports.Finally, the θ when internal state value U is greater than neuronic threshold value _ijtime, neuron sends pulse, i.e. Y _ij=1.In iterative process, threshold value θ _ijdo the change of nonlinear exponential damping, damped expoential is α ^θ, but after sending pulse, θ _ijan amplitude coefficient has also been superposed while carrying out exponential damping .

C, image Laplacian pyramid, the conversion of the laplacian pyramid of image is obtaining by the basis of gaussian pyramid.Therefore, Laplacian pyramid is divided into two steps: first image is carried out gaussian pyramid decomposition, then obtain laplacian pyramid.Gauss operator and original image (representing with G0) are carried out convolution (i.e. Gassian low-pass filter), then the down-sampling of interlacing every row is carried out, obtain image being similar on low resolution, resolution is the half (ground floor of gaussian pyramid) of former figure.Again low-pass filtering and down-sampling are entered to the image after sampling, obtain next layer of gaussian pyramid.So repeatedly carry out aforesaid operations, obtain some layers, in order to form gaussian pyramid.Be 1/4 of last tomographic image by this tomographic image size of gaussian pyramid obtained above, then utilize method of interpolation to carry out interpolation expansion to gaussian pyramid, make l tomographic image G _l, the size after expansion and l-1 tomographic image G _l-1measure-alike, its account form is as follows:

Wherein,

Order

Wherein, in above formula, w (m, n) is low pass window function, and size is 5 × 5, LP _lfor laplacian pyramid l layer, LP _nfor laplacian pyramid n-th layer.

D, rebuild original image by laplacian pyramid; Each tomographic image of laplacian pyramid is amplified to the same with next tomographic image large through progressively interpolation, and then is added and can rebuilds original image; When the number of plies of laplacian pyramid arrives N 0, its energy size can reflect the quality of image, and the mode of all employing region energies chooses coefficient.

S-PCNN is to the information sensing such as details, edge of image, and the high level of laplacian pyramid contains details and the texture variations information of image, so adopt S-PCNN model of the present invention to choose to retain details and the texture information of image in high layer coefficients.S, V component is carried out Laplacian pyramid respectively, and by above-mentioned strategy, each tomographic image information is synthesized new S, V component.By the new HSV coloured image obtained, be converted to rgb space, obtain the coloured image after merging.As can be seen from Table 1, algorithm other algorithms of numeric ratio in these evaluation objective indicators are comparatively effective herein, wherein, SF expression space frequency, STD, expression standard deviation, EN represents entropy, AV represents average gradient, and M represents average, and these values are all that larger expression picture quality is better.Can find out in Fig. 2 that the present invention is the most clear, and color and source images are the most close, demonstrate validity and the feasibility of color image fusion herein.

Table 1. is based on the fusion mass evaluation of estimate of different fusion method gained fused images.

	SF	STD	EN	AV	M
						Weighting	12.57	65.01	7.45	4.20	116.97
PCA	12.76	65.17	7.44	4.23	166.67
						WT	13.19	65.39	7.49	4.31	163.33
PCNN	17.62	66.27	7.45	5.27	168.50
						Herein	20.40	67.08	7.44	5.85	166.96

The present invention is not limited to above-mentioned preferred forms; anyone can draw other various forms of products under enlightenment of the present invention; no matter but any change is done in its shape or structure; every have identical with the application or akin technical scheme, all drops within protection scope of the present invention.

Claims (1)

1., based on a Color Image Fusion method for S-PCNN and laplacian pyramid, it is characterized in that: specifically comprise the following step:

A. first, by RGB color image space transforming good for registration to HSV color space;

B., after utilizing S-PCNN to carry out characteristic area cluster to H component, merge with the H component realizing each source images based on pulse oscillating frequency figure; H component processing procedure: first H component is sent into S-PCNN model and carry out iteration, obtain igniting and frequently scheme OFG, S-PCNN igniting corresponding to H component is more frequently schemed OFG and is carried out local entropy computing, obtain the eigenmatrix of OFG local entropy matrix as H component, finally contrast the size of the H component respective pixel local entropy LE of different source images, get the pixel with larger local entropy as fusion pixel;

C. image Laplacian pyramid, wherein, S, V component, utilizes laplacian pyramid to carry out resolution decomposition S, V component, then utilizes different fusion rule to merge S, V component;

D. last the inverse transformation of HSV color space is carried out to new H, S, V component that steps A, B, C obtain, finally obtain the RGB color image after merging.