基本的FreeMat / MATLAB语法 - 维度错误

Question

我正在使用FreeMat，我有一张RGB图片，这是一张3D矩阵，包含图片的列和行以及每个像素的RGB值。

由于没有将RGB图片转换为YIQ的内在函数，我实现了一个。我想出了这段代码：

假设我有一个3D数组，image_rgb：

matrix = [0.299 0.587 0.114;
0.596 -0.274 -0.322;
0.211 -0.523 0.312];
row = 1:length(image_rgb(:,1,1));
col = 1:length(image_rgb(1,:,1));
p = image_rgb(row,col,:);

%Here I have the problem
mage_yiq(row,col,:) = matrix*image_rgb(row,col,:);

max_y = max (max(image_yiq(:,:,1)));
max_i = max (max(image_yiq(:,:,2)));
max_q = max (max(image_yiq(:,:,3)));

%Renormalize the image again after the multipication
% to [0,1].
image_yiq(:,:,1) = image_yiq(:,:,1)/max_y;
image_yiq(:,:,2) = image_yiq(:,:,2)/max_i;
image_yiq(:,:,3) = image_yiq(:,:,3)/max_q;

我无法理解为什么矩阵乘法失败。我希望代码很好而不仅仅是，手工繁殖矩阵......

Answer 1

您尝试使用您创建的matrix多个3D数组，这不是正确的矩阵乘法。您应该将图像数据展开为3×m * n矩阵，并将其与自定义矩阵相乘。

以下是将自定义色彩空间转换应用于RGB图像的解决方案。我使用了您提供的矩阵，并将其与内置的YIQ变换进行了比较。

%# Define the conversion matrix
matrix = [0.299  0.587  0.114;
          0.596 -0.274 -0.322;
          0.211 -0.523  0.312];

%# Read your image here
rgb = im2double(imread('peppers.png'));
subplot(1,3,1), imshow(rgb)
title('RGB')


%# Convert using unfolding and folding
[m n k] = size(rgb);

%# Unfold the 3D array to 3-by-m*n matrix
A = permute(rgb, [3 1 2]);
A = reshape(A, [k m*n]);

%# Apply the transform
yiq = matrix * A;

%# Ensure the bounds
yiq(yiq > 1) = 1;
yiq(yiq < 0) = 0;

%# Fold the matrix to a 3D array
yiq = reshape(yiq, [k m n]);
yiq = permute(yiq, [2 3 1]);

subplot(1,3,2), imshow(yiq)
title('YIQ (with custom matrix)')


%# Convert using the rgb2ntsc method
yiq2 = rgb2ntsc(rgb);
subplot(1,3,3), imshow(yiq2)
title('YIQ (built-in)')

请注意，k对于RGB图像为3。在每个语句后查看矩阵的大小。不要忘记将图片转换为double。

Answer 2

使用相同的矩阵和-color-matrix函数可以使用Imagemagick执行此操作：

输入：

--Start Script

sendToAHK = function (key)

    --print('It was assigned string:    ' .. key)

    local file = io.open("C:\\Users\\TaranWORK\\Documents\\GitHub\\2nd-keyboard-master\\LUAMACROS\\keypressed.txt", "w") -- writing this string to a text file on disk is probably NOT the best method. Feel free to program something better!

    --Make sure to substitute the path that leads to your own "keypressed.txt" file, using the double backslashes.

    --print("we are inside the text file")

    file:write(key)

    file:flush() --"flush" means "save"

    file:close()

    lmc_send_keys('{F24}')  -- This presses F24. Using the F24 key to trigger AutoHotKey is probably NOT the best method. Feel free to program something better!

end



local config = { -- this is line 85

    [45]  = "insert",

    [36]  = "home",

    [33]  = "pageup",

    [46]  = "delete",

    [35]  = "end",

    [34]  = "pagedown",

    [27]  = "escape",

    [112] = "F1",

    [113] = "F2",

    [114] = "F3",

    [115] = "F4",

    [116] = "F5",

    [117] = "F6",

    [118] = "F7",

    [119] = "F8",

    [120] = "F9",

    [121] = "F10",

    [122] = "F11",

    [123] = "F12",

    [8]   = "backspace",

    [220] = "backslash",

    [13]  = "enter",

    [16]  = "rShift",

    [17]  = "rCtrl",

    [38]  = "up",

    [37]  = "left",

    [40]  = "down",

    [39]  = "right",

    [32]  = "space",

    [186] = "semicolon",

    [222] = "singlequote",

    [190] = "period",

    [191] = "slash",

    [188] = "comma",

    [219] = "leftbracket",

    [221] = "rightbracket",

    [189] = "minus",

    [187] = "equals",

    [96]  = "num0",

    [97]  = "num1",

    [98]  = "num2",

    [99]  = "num3",

    [100] = "num4",

    [101] = "num5",

    [102] = "num6",

    [103] = "num7",

    [104] = "num8",

    [105] = "num9",



    [106] = "numMult",

    [107] = "numPlus",

    [108] = "numEnter" --sometimes this is different, check your keyboard

    [109] = "numMinus",

    [110] = "numDelete",

    [111] = "numDiv",

    [144] = "numLock", --probably it is best to avoid this key. I keep numlock ON, or it has unexpected effects



    [192] = "`",  --this is the tilde key just before the number row

    [9]   = "tab",

    [20]  = "capslock",

    [18]  = "alt",





    [string.byte('Q')] = "q",

    [string.byte('W')] = "w",

    [string.byte('E')] = "e",

    [string.byte('R')] = "r",

    [string.byte('T')] = "t",

    [string.byte('Y')] = "y",

    [string.byte('U')] = "u",

    [string.byte('I')] = "i",

    [string.byte('O')] = "o",

    [string.byte('P')] = "p",

    [string.byte('A')] = "a",

    [string.byte('S')] = "s",

    [string.byte('D')] = "d",

    [string.byte('F')] = "f",

    [string.byte('G')] = "g",

    [string.byte('H')] = "h",

    [string.byte('J')] = "j",

    [string.byte('K')] = "k",

    [string.byte('L')] = "l",

    [string.byte('Z')] = "z",

    [string.byte('X')] = "x",

    [string.byte('C')] = "c",

    [string.byte('V')] = "v",

    [string.byte('B')] = "b",

    [string.byte('N')] = "n",

    [string.byte('M')] = "m",



    [string.byte('0')] = "0",

    [string.byte('1')] = "1",

    [string.byte('2')] = "2",

    [string.byte('3')] = "3",

    [string.byte('4')] = "4",

    [string.byte('5')] = "5",

    [string.byte('6')] = "6",

    [string.byte('7')] = "7",

    [string.byte('8')] = "8",

    [string.byte('9')] = "9",



    --[255] = "printscreen" --these keys do not work

}



    -- define callback for whole device

lmc_set_handler('MACROS', function(button, direction)

    --Ignoring upstrokes ensures keystrokes are not registered twice, but activates faster than ignoring downstrokes. It also allows press and hold behaviour

    if (direction == 0) then return end -- ignore key upstrokes.

    if type(config[button]) == "string" then

        print(' ')

        print('Your key ID number is:   ' .. button)

        print('It was assigned string:    ' .. config[button])

        sendToAHK(config[button])

    else

        print(' ')

        print('Not yet assigned: ' .. button)

    end

end)

但这不是真正的sRGB转YIQ。

这是一个sRGB到YIQ转换，显示YIQ就好像RGB：

convert peppers_tiny.png -color-matrix \
" \
0.299 0.587 0.114 \
0.596 -0.274 -0.322 \
0.211 -0.523 0.312 \
" \
peppers_tiny_yiq.png

这是相同的，但交换前两个频道：

convert peppers_tiny.png -colorspace YIQ -separate \
-set colorspace sRGB -combine peppers_tiny_yiq2.png