how to convert RGB to HSV(Lossless) in matlab?

https://bitbucket.org/chernov/colormath/src/012a5ae28033dc8e91ec66ec349c95d81caddb71/colormath/hsv.cpp?at=default&fileviewer=file-view-default. This link gave me code to convert rgb to hsv lossless but it is in C++ . I want in matlab. I am new to matlab please help me.

http://www.sciencedirect.com/science/article/pii/S0045790615002827. This is the paper which I am referring.

Answer 1

[Walter Roberson]

rgb2hsv(im2double(YourImage))

Note: this is only lossless if YourImage is no more than approximately 32 bit depth (I would need to work out the exact boundaries.) In particular, if your image is double precision to start with and uses the full 53 bits of precision per channel per pixel, then this will not be lossless; if you have those kinds of precision requirements then you will need to use an extended precision mechanism (such as the Symbolic Toolbox)

[Aarach Sap]

Thank you for the reply. But I am doing reversible data hiding. Therefore, I don't want to use rgb2hsv built-in function. I want to implement the code give above(C++) in matlab. But I find difficulty in writing in matlab.

 % % #include "hsv.h"
% % 
% % namespace colormath {
% % 
% % void rgb2hsv(uint8_t r, uint8_t g, uint8_t b, uint32_t& h, uint16_t& s, uint8_t& v) {
% %     uint8_t max, mid, min;
% %     uint32_t edge;
% %     bool inverse;
% % 
% %     if (r > g) {
% %         if (g >= b) {
% %             max = r;
% %             mid = g;
% %             min = b;
% %             edge = 0;
% %             inverse = false;
% %         } else if (r > b) {
% %             max = r;
% %             mid = b;
% %             min = g;
% %             edge = 5;
% %             inverse = true;
% %         } else {
% %             max = b;
% %             mid = r;
% %             min = g;
% %             edge = 4;
% %             inverse = false;
% %         }
% %     } else if (r > b) {
% %         max = g;
% %         mid = r;
% %         min = b;
% %         edge = 1;
% %         inverse = true;
% %     } else {
% %         if (g > b) {
% %             max = g;
% %             mid = b;
% %             min = r;
% %             edge = 2;
% %             inverse = false;
% %         } else {
% %             max = b;
% %             mid = g;
% %             min = r;
% %             edge = 3;
% %             inverse = true;
% %         }
% %     }
% % 
% %     v = max;
% % 
% %     uint32_t delta = max - min;
% %     if (delta == 0) {
% %         s = 0;
% %         h = 0;
% %         return;
% %     }
% % 
% %     s = ((delta << 16) - 1) / v;
% % 
% %     h = (((mid - min) << 16) / delta) + 1;
% %     if (inverse)
% %         h = hueEdgeLen - h;
% %     h += edge * hueEdgeLen;
% % }
% % 
% % void hsv2rgb(uint32_t h, uint16_t s, uint8_t v, uint8_t& r, uint8_t& g, uint8_t& b) {
% %     if (s == 0 || v == 0) {
% %         r = v;
% %         g = v;
% %         b = v;
% %         return;
% %     }
% % 
% %     uint32_t delta = ((s * v) >> 16) + 1;
% %     uint8_t min = v - delta;
% %     uint8_t* mid;
% % 
% %     if (h >= hueEdgeLen * 4) {
% %         h -= hueEdgeLen * 4;
% %         if (h < hueEdgeLen) {
% %             b = v;
% %             g = min;
% %             mid = &r;
% %         } else {
% %             h -= hueEdgeLen;
% %             h = hueEdgeLen - h;
% %             r = v;
% %             g = min;
% %             mid = &b;
% %         }
% %     } else if (h >= hueEdgeLen * 2) {
% %         h -= hueEdgeLen * 2;
% %         if (h < hueEdgeLen) {
% %             g = v;
% %             r = min;
% %             mid = &b;
% %         } else {
% %             h -= hueEdgeLen;
% %             h = hueEdgeLen - h;
% %             b = v;
% %             r = min;
% %             mid = &g;
% %         }
% %     } else {
% %         if (h < hueEdgeLen) {
% %             r = v;
% %             b = min;
% %             mid = &g;
% %         } else {
% %             h -= hueEdgeLen;
% %             h = hueEdgeLen - h;
% %             g = v;
% %             b = min;
% %             mid = &r;
% %         }
% %     }
% % 
% %     *mid = ((h * delta) >> 16) + min;
% % }
% % 
% % } // namespace colormath

[Walter Roberson]

Whatever. Go ahead and implement that code in MATLAB then. Remember you will need to loop.

(delta << 16)

is delta * 2^16

(h * delta) >> 16

is

(h * delta) / 2^16

and

h -= hueEdgeLen * 2

is

h = h - hueEdgeLen * 2

For the bits about

mid = &g

use mid = 1 instead of &r, mid = 2 instead of &g, mid = 3 instead of &b, and then at the end replace

*mid = ((h * delta) >> 16) + min;

with

if mid == 1
  r = ((h * delta) / 2^16) + min;
elseif mid == 2
  g = ((h * delta) / 2^16) + min;
else
  b = ((h * delta) / 2^16) + min;
end

[Aarach Sap]

Thank you. I tried to implement the code. But there are errors. One of them is on 'bool inverse'. Here is the code which I did.

RGB=imread('Lenna.tiff');
r=RGB(:,:,1);
g=RGB(:,:,2);
b=RGB(:,:,3);
 bool inverse;
%  maxValue = 0XFF;
%  maxSaturation = 0XFFFF;
 hueEdgeLen = 65537;
 maxHue = hueEdgeLen * 6;
  if (r > g) 
        if (g >= b) 
            max = r;
            mid = g;
            min = b;
            edge = 0;
            inverse = false;

           elseif (r > b)
              max = r;
              mid = b;
              min = g;
              edge = 5;
              inverse = true;

              else
              max = b;
              mid = r;
              min = g;
              edge = 4;
              inverse = false;
              end

    elseif (r > b)
          max = g;
          mid = r;
          min = b;
          edge = 1;
          inverse = true;
       else 
          if (g > b) 
              max = g;
              mid = b;
              min = r;
              edge = 2;
              inverse = false;

          else

              max = b;
              mid = g;
              min = r;
              edge = 3;
              inverse = true;
          end
          end

      v = max;

      delta = max - min;
      if (delta == 0) 
          s = 0;
          h = 0;
          return;
      end
  %      s = ((delta < 16) - 1)./ v;
       s = ((delta.(2^16)) - 1)./ v;
      h = (((mid - min) < 16)./ delta) + 1;
     if (inverse)
       h = hueEdgeLen - h;
       h = h + (edge * hueEdgeLen);
     end

[Walter Roberson]

Replace

bool inverse;

by

inverse = false;

[Aarach Sap]

I replaced

h = (((mid - min) < 16)./ delta) + 1;

with

h = (((mid - min)* (2^ 16))./ delta) + 1;

This line worked. I run following code.

There is no error. But I am confused why the elements of 'delta and s' are all zero.

RGB=imread('Lenna.tiff');
r=RGB(:,:,1);
g=RGB(:,:,2);
b=RGB(:,:,3);
 inverse = false;
%  maxValue = 0XFF;
%  maxSaturation = 0XFFFF;
 hueEdgeLen = 65537;
 maxHue = hueEdgeLen * 6;
  if (r > g) 
        if (g >= b) 
            max = r;
            mid = g;
            min = b;
            edge = 0;
            inverse = false;

           elseif (r > b)
              max = r;
              mid = b;
              min = g;
              edge = 5;
              inverse = true;

              else
              max = b;
              mid = r;
              min = g;
              edge = 4;
              inverse = false;
              end

    elseif (r > b)
          max = g;
          mid = r;
          min = b;
          edge = 1;
          inverse = true;
       else 
          if (g > b) 
              max = g;
              mid = b;
              min = r;
              edge = 2;
              inverse = false;

          else

              max = b;
              mid = g;
              min = r;
              edge = 3;
              inverse = true;
          end
          end

      v = max;

      delta = max - min;
      if (delta == 0) 
          s = 0;
          h = 0;
          return;
      end
  %      s = ((delta < 16) - 1)./ v;
       s = ((delta*(2^16)) - 1)./ v;
      h = (((mid - min)* (2^ 16))./ delta) + 1;
     if (inverse)
       h = hueEdgeLen - h;
       h = h + (edge * hueEdgeLen);
     end

     HSVimage=cat(3,h,s,v);
     imshow(HSVimage);

[Walter Roberson]

Your data might be uint8. Subtracting a larger uint8 from a smaller gives 0. In your C code there was probably an implicit conversion to signed or double.

[Aarach Sap]

Thank you. Now I use double and run program. I did RGB to HSV image and try to recover HSV to RGB . But why it is not working. Below is my code.

%rgb to HSv
RGB=cast(imread('Lenna.tiff'),'double');
% RGB=imread('Lenna.tiff');
r=RGB(:,:,1);
g=RGB(:,:,2);
b=RGB(:,:,3);
 inverse = false;
%  maxValue = 0XFF;
%  maxSaturation = 0XFFFF;
 hueEdgeLen = 65537;
 maxHue = hueEdgeLen * 6;
  if (r > g) 
        if (g >= b) 
            max = r;
            mid = g;
            min = b;
            edge = 0;
            inverse = false;

           elseif (r > b)
              max = r;
              mid = b;
              min = g;
              edge = 5;
              inverse = true;

              else
              max = b;
              mid = r;
              min = g;
              edge = 4;
              inverse = false;
              end

    elseif (r > b)
          max = g;
          mid = r;
          min = b;
          edge = 1;
          inverse = true;
       else 
          if (g > b) 
              max = g;
              mid = b;
              min = r;
              edge = 2;
              inverse = false;

          else

              max = b;
              mid = g;
              min = r;
              edge = 3;
              inverse = true;
          end
          end

      v = max;

      delta = max - min;
      if (delta == 0) 
          s = 0;
          h = 0;
          return;
      end
  %      s = ((delta < 16) - 1)./ v;
       s = ((delta*(2^16)) - 1)./ v;
      h = (((mid - min)* (2^ 16))./ delta) + 1;
     if (inverse)
       h = hueEdgeLen - h;
       h = h + (edge * hueEdgeLen);
     end

     HSVimage=cat(3,h,s,v);
      HSVdisplay=cast(HSVimage,'uint8');
     figure(1);
     imshow(HSVdisplay);
     imwrite(HSVdisplay,'rgb2hsvimage.tiff');

%   % HSV to RGB 
HSV=cast(imread('rgb2hsvimage.tiff'),'double');
% HSV=imread('rgb2hsvimage.tiff');
h=HSV(:,:,1);
s=HSV(:,:,2);
v=HSV(:,:,3);
 hueEdgeLen = 65537;

      if (s == 0 | v == 0) 
          r = v;
          g = v;
          b = v;
          return;
      end
   delta = ((s * v)/(2^16)) + 1;
   min = v - delta;
  % %     uint8_t* mid;

      if (h >= hueEdgeLen * 4) 
          h =h- hueEdgeLen * 4;
          if (h < hueEdgeLen) 
               b = v;
               g = min;
             mid = 1;

            else 
              h =h- hueEdgeLen;
               h = hueEdgeLen - h;
              r = v;
               g = min;
              mid = 2;
          end

      elseif (h >= hueEdgeLen * 2) 
           h =h- hueEdgeLen * 2;
          if (h < hueEdgeLen) 
               g = v;
               r = min;
              mid = 2;

       else
              h =h- hueEdgeLen;
               h = hueEdgeLen - h;
              b = v;
              r = min;
               mid = 2;

          end

       else 
           if (h < hueEdgeLen) 
               r = v;
               b = min;
              mid = 2;

           else
               h =h- hueEdgeLen;
               h = hueEdgeLen - h;
               g = v;
              b = min;
              mid = 1;

           end
      end

     % %     *mid = ((h * delta) >> 16) + min;         

     if mid == 1
    r = ((h * delta) / 2^16) + min;
  elseif mid == 2
    g = ((h * delta) / 2^16) + min;
  else
    b = ((h * delta) / 2^16) + min;
  end
         RGBimage=cat(3,r,g,b);
         RGBdisplay=cast(RGBimage,'uint8');
        figure(2);
        imshow(RGBdisplay);

[Walter Roberson]

As I wrote above:

Remember you will need to loop.

The C++ code you show is for converting one pixel. To use it you will need to loop over all pixels in your image.

Please remind us why using the Mathworks-supplied rgb2hsv() routine will not work for you? Have you found different RGB->HSV conversion equations than MATLAB implements?

[Aarach Sap]

In fact, I want to convert rgb to hsv and then perform some operation but on the integer values. Finally, want to reverse back to original without any loss.

[Guillaume]

Matlab is not C. Writing matlab code like it was C code is not going to be efficient. You could write an efficient and accurate matlab rgb to hsv conversion in about 5 lines of code. However, mathworks have already done the work for you with the rgb2hsv and hsv2rgb function they provide.

The code you've written will not be more accurate than the function provided by matlab and certainly will be a lot slower, so what's the point?

[Walter Roberson]

You are presuming here that lossless rgb to hsv to rgb is possible when you use uint8 to store the hsv . It is not clear that is the case.

If you use

HSVd = rgb2hsv(YourImage);
HSVu = im2uint8(HSVd);

HSVud = im2double(HSVu);
RGBd = hsv2rgb(HSVud);
RGBu = im2uint8(RGBd);

then HSVu will be a uint8 representation of the HSV, and RGBu will be the reconstruction after going through the RGB->HSV (uint8) -> RGB steps. You will find that the values are not exactly the same. My tests show a maximum difference of 3.

[gR,gG,gB] = ndgrid(uint8(0:255));
YourImage = cat(3, reshape(gR,[],size(gR,3)), reshape(gG,[],size(gG,3)), reshape(gB, [], size(gB,3)));
HSVd = rgb2hsv(YourImage);
HSVu = im2uint8(HSVd);
HSVud = im2double(HSVu);
RGBd = hsv2rgb(HSVud);
RGBu = im2uint8(RGBd);
max(max(abs(double(RGBu)-double(YourImage))))

However if you do not go through uint8,

HSVd = rgb2hsv(YourImage);
RGBd = hsv2rgb(HSVd);
RGBu = im2uint8(RGBd);

then the conversion is exact.

You could try other methods of converting from the [0,1] range to [0,255], such as floor(HSVd*255) or round() or fix(), but you will find that those then fail the test of converting exactly when converting im2double(YourImage) back to YourImage, which im2uint8() handles fine.

Answer 2

[Dilshad Ahmad]

You can do by colormap function like I=Imread('your colour image');

Colormap HSV

[Walter Roberson]

Unfortunately not relevant to this question.