Image processing | RGB and HSI conversion and filtering

1. Convert RGB image to HSI image

Specific implementation steps:

1. Convert the input image to double type by im2double() function and normalize it;
2. Double loop to traverse each pixel of the image: obtain the three components of R, G, and B respectively, and bring in the conversion formula to calculate and solve. Calculated as follows:

1) Function interface realization:

function HSI = myRGB2HSI(RGB)
%   从RGB颜色空间向HSI颜色空间的转换
%   RGB(uint8):输入的RGB彩色图像
%   HSI(double):转换后的HSI彩色图像
img1_double = im2double(RGB);   % 转成double并作归一化处理   
[r, c, k] = size(img1_double);
H = zeros(r, c);
S = zeros(r, c);
I = zeros(r, c);

for i = 1 : r
    for j = 1 : c
        % 分别获取R,G,B分量
        R = img1_double(i,j,1);
        G = img1_double(i,j,2);
        B = img1_double(i,j,3);
        fenzi = 0.5 * ( (R-G)+(R-B) );
        fenmu = sqrt( (R-G)^2 + (R-B)*(G-B) );
        % 易错点：分母需加上eps防止为0
        xita = acos( fenzi/(fenmu+eps) );
        if ( B<=G )
            HSI(i,j,1) = xita;
        else
            HSI(i,j,1) = 2*pi-xita;    
        end 
        HSI(i,j,1) = HSI(i,j,1) / (2*pi);  % H分量需要除以2*pi进行归一化
        min_value = min(min(R,G),B);
        % 易错点：分母需加上eps防止为0
        HSI(i,j,2) = 1 - ( 3/(R+G+B+eps) ) * min_value;
        HSI(i,j,3) = (R+G+B)/3;
    end
end
end

2) Note:

1. The angle and the radian should correspond to each other before and after, either using the angle or all using the radian;
2. Add eps to the denominator to prevent it from being 0;
3. Keep awake about the range of input and output data at all times, that is, do a good job of normalization and scaling;

3) Processing result:

2. Convert HSI image to RGB image

Specific implementation steps:

1. Convert the input image to double type by im2double() function and normalize it;
2. Traverse each pixel, first multiply the H component by 2*pi to expand it, and then calculate the corresponding R, G, and B components according to the formula according to the range of H. The calculation formula is as follows:

1) Function interface realization:

function RGB = myHSI2RGB(HSI)
% 从HSI颜色空间向RGB颜色空间的转换
% HSI(double):输入的HSI彩色图像
% RGB(uint8): 转换后的RGB彩色图像
HSI = im2double(HSI);
[r,c,k] = size(HSI);
RGB = zeros(r,c,k);
for i = 1 : r
    for j = 1 : c
        H = HSI(i,j,1)*2*pi;
        S = HSI(i,j,2);
        I = HSI(i,j,3);
        if ( H>=0 && H<2/3*pi)
            expression = S*cos( H )/(cos( pi/3-H ) + eps);
            RGB(i,j,1) = I * ( 1+expression );
            RGB(i,j,3) = I * (1-S);
            RGB(i,j,2) = 3*I - ( RGB(i,j,1)+RGB(i,j,3) );
        elseif ( H>=2/3*pi && H<4/3*pi)
            H = H-2*pi/3;
            RGB(i,j,1) = I * (1-S);
            expression = S*cos( H )/(cos( pi/3-H ) + eps);
            RGB(i,j,2) = I * ( 1+ expression );
            RGB(i,j,3) = 3*I - ( RGB(i,j,1)+RGB(i,j,2) );
        elseif (H>=4/3*pi && H<=2*pi)
            H = H-4*pi/3;
            RGB(i,j,2) = I * (1-S);
            expression = S*cos( H )/(cos( pi/3-H ) + eps);
            RGB(i,j,3) = I * ( 1+ expression );    
            RGB(i,j,1) = 3*I - ( RGB(i,j,2)+RGB(i,j,3) );
        end
    end
end
RGB = RGB * 255;
RGB = uint8(RGB);
end

2) Note:

1. The angle and the radian should correspond to each other before and after, either using the angle or all using the radian;
2. Add eps to the denominator to prevent it from being 0;
3. Keep awake about the range of input and output data at all times, that is, do a good job of normalization and scaling;

3) Processing result:

3. Perform mean filtering in HSI space

Specific implementation steps:

1. Convert RGB image to hsi space;
2. Perform average filtering on the luminance component I in the HSI color space;
3. Convert the average filtering result to RGB space;

Note: When extracting the I component, first multiply it by 255 to expand the gray level. After the filtering process is completed, divide it by 255 and assign it to the third component of the HSI image. The filter function needs to use im2double to read in the image.

1) Main function call:

clc;
clear all;

img1 = imread('images/EXP5.tif');
subplot(1,2,1);
imshow(img1);
title('输入的 RGB 图像','FontSize',20,'FontName','微软雅黑');


% 1.将rbg图像转换到hsi空间
img_hsi = myRGB2HSI(img1);
% 2.对HSI颜色空间中的亮度分量I进行滤波处理
w = [1,2,3,2,1; 2,5,6,5,2; 3,6,8,6,3; 2,5,6,5,2; 1,2,3,2,1];
i = img_hsi(:,:,3) * 255;
temp = mySpatialFilter(i,[5,5],w);
temp = temp/255;
% 3.将滤波后的结果转回至RGB空间中
img_hsi(:,:,3) = temp;

img2 = myHSI2RGB(img_hsi);
subplot(1,2,2);
imshow(img2);
title('在HSI空间滤波后的图像','FontSize',20,'FontName','微软雅黑');

2) Mean filter function interface realization:

function img_enhanced = mySpatialFilter( img1 ,fsize, w )
% 均值滤波函数
% img1：输入的rgb图像；img_enhanced：输出图像；fsize：滤波器掩模尺寸
% w：滤波器系数

img1_double = im2double(img1);
[r c] = size(img1_double);
% 获取滤波器模板尺寸
m = fsize(1);
n = fsize(2);
% 补边后的总边数
row_fill = r + m - 1;
col_fill = c + n - 1;
% 用 0 填充边界
img2 = zeros(row_fill, col_fill);
img2(1+(m-1)/2:row_fill-(m-1)/2,1+(n-1)/2:col_fill-(n-1)/2) = img1_double;
img2_copy = img2;
for i = 1 : r
    for j = 1 : c
        % 提取与滤波器模板等大的区域
        img_area = img2_copy(i:i+m-1,j:j+n-1);
        x_center = i + (m-1)/2;
        y_center = j + (n-1)/2;
        % 计算滤波处理后的图像
        img2(x_center,y_center) = sum(w(:).*img_area(:))/sum(w(:));
    end
end
img_enhanced= img2(1+(m-1)/2:row_fill-(m-1)/2,1+(n-1)/2:col_fill-(n-1)/2);
end

3) Note:

1. The mean filter function needs to use im2double to read in the image;

4) Processing result: