[image registration] improved SAR image registration based on sar-sift with matlab code

1 brief introduction

Compared to traditional  SIFT  Algorithm ,SAR-SIFT  It is mainly modified in two aspects , The first is to improve the pixel gradient calculation method , The difference is replaced by the ratio . The second is to improve the method of generating the final descriptor . Through the modification of these two aspects , This makes this method more suitable for general multiplicative speckle noise  SAR  Images . about  SAR  Multiplicative noise in images ,SAR-SIFT  To some extent, compared with the original  SIFT  Better adaptability , But when  SAR  When there are strong noise spots in the image ,SAR-SIFT  There will also be a large number of outliers when the algorithm extracts feature points . in the light of  SAR  Strong multiplicative speckle noise in pictures , First of all, with the help of SAR-SIFT  The algorithm extracts features , Then, sparse coding is used to effectively use feature clustering to eliminate the outliers caused by strong speckle noise in the process of feature extraction . Thus through  SAR-SIFT  Algorithm combination FSC  Algorithm to suppress  SAR  Noise spots in the image , Improve  SAR  The accuracy and robustness of image registration .​

2 Part of the code

function [k_percentile]=compute_k_percentile_auto(gradient_x,gradient_y,perc)

% This function calculates a contrast parameter k, This contrast parameter is used to calculate the diffusion coefficient

%gradient_x Is the gradient in the horizontal direction ,gradient_y Is the gradient in the vertical direction

%perc Is the percentile of the gradient histogram , The default value is 0.7,k The value of is determined according to the percentile

% The conductivity functions are all k The increasing function of , So for the same gradient value , If k It's worth more , Then the value of conduction coefficient is larger

% Therefore, the diffusion is large , Severe smoothness , So we can see that , If you want to keep the details, you need smaller k value

% This function automatically calculates k value , So there is no need to specify bin Size

% Histogram interval

unit=0.005;

% Ignore the boundary to calculate the maximum value of the gradient

gradient=sqrt(gradient_x.^2+gradient_y.^2);

[M,N]=size(gradient);

temp_gradient=gradient(2:M-1,2:N-1);

% Ignore Boundary Calculation histogram

temp_gradient=temp_gradient(temp_gradient>0);

max_gradient=max(max(temp_gradient));

min_gradient=min(min(temp_gradient));

temp_gradient=round((temp_gradient-min_gradient)/unit);

nbin=round((max_gradient-min_gradient)/unit);

hist=zeros(1,nbin+1);

[M1,N1]=size(temp_gradient);

sum_pix=M1*N1;% Number of non-zero pixel gradients

% Calculate the histogram

for i=1:1:M1

    for j=1:1:N1

        hist(temp_gradient(i,j)+1)=hist(temp_gradient(i,j)+1)+1;

    end

end

% Histogram percentile

nthreshold=perc*sum_pix;

nelements=0;

temp_i=0;

for i=1:1:nbin+1

    nelements=nelements+hist(i);

    if(nelements>=nthreshold)

        temp_i=i;

        break;

    end

end

%   k_percentile=max_gradient*(temp_i)/(nbin+1);

    k_percentile=(temp_i-1)*unit+min_gradient;

end

3 Simulation results

4 reference

[1] Li Pei , Jiang Gang , and Ma Qianli . " be based on SAR-SIFT The improved SAR Image registration ." Mapping Bulletin (2021).

About bloggers ： Good at intelligent optimization algorithms 、 Neural networks predict 、 signal processing 、 Cellular automata 、 The image processing 、 Path planning 、 UAV and other fields Matlab Simulation , relevant matlab Code problems can be exchanged by private letter .