最近在用octave (類似於matlab的計算軟體， 函式和matlab一致) 寫程式的時候， 在顯示影象和儲存影象的時候遇到了一些小問題， 所以簡單的總結了一下。

本文用的影象為灰度影象：

imread() 返回的影象型別是uint8型別， 這時用imshow顯示影象的時候， imshow會認為輸入矩陣的範圍在0-255， 如果imshow的引數為double型別的，那麼imshow認為輸入矩陣的值為0-1.

很多時候需要將影象轉換為double型別的， 但是轉換以後直接使用imshow顯示的是一片白色， 是因為當imshow顯示影象的時候， 會認為double型別的影象矩陣的範圍在0-1， 超過1的畫素值當作1處理， 這樣就是幾乎所有的畫素都是白色。

情況1：

% img will be uint8 type
img = imread('syz.bmp');


% imshow: when the parameter img is uint8 type, 
% imshow will default make the range of pixel value as [0-255]
imshow(img);
fprintf('Program paused. Press enter to continue.\n');
pause;

顯示結果：

情況2：

當把影象轉換double資料型別後：

% sometimes we need to process the img by double type,
% so, you may convert the data type to double
dimg = double(img);

% but, right now you will not get the correct display,
% because if the parameter of imshow is double type, if will defaultly 
% take range of [0-1], but now the range is [0-255]
% all the value over 1 is ceilled to 1. (clamped to 1)
% so, the displayed image will be a whole white picture.
imshow(dimg);
fprintf('Program paused. Press enter to continue.\n');
pause;
 

這個時候， 因為imshow的引數為double型別， imshow認為引數的值為0-1， 但是實際還是0-255， 這樣超過1的值被認為是1， 顯示幾乎全是白色

情況3： 如何正確的顯示 double 型別的影象呢， 有兩種方式

% how to correctly display double typed image
% way 1: convert double to uint8
imshow(uint8(dimg));
fprintf('Program paused. Press enter to continue.\n');
pause;

% way 2: change the value of double typed image to [0-1]
maxVal = max(max(dimg));
imshow(dimg / maxVal);
fprintf('Program paused. Press enter to continue.\n');
pause;

顯示結果：

這兩個顯示結果是一樣的。

情況4： 有些時候，影象經過處理以後會存在值為負數的畫素， 如何顯示呢？

這需要將所有的負數處理掉， 可以將所有的畫素減去這個影象的最小的畫素值， 最小的畫素值為負數， 那麼影象的畫素值就都為正數了。

%% some other occurence, the image maybe in double type, 
% but some normalization operation will lead to negative pixel value
% In order to display the image, we need to add a value that make
% all negative value to positive value

% here i just minus a specific value, in real application,
% a face image maybe normalized by substract mean face.
normImg = dimg - 127; % then normImg will have some negative values

minVal = min(min(normImg));
imshow(uint8(normImg - minVal));
fprintf('Program paused. Press enter to continue.\n');
pause;

下面是整個示例程式碼：

%% this file is used to show how to use imshow or imagesc and so on
% there is some different when you use imshow to display double image or uint8 image
% here all the code will process gray image, RGB color image will not suitable for this file's code
% but you can convert RGB color image to gray image if you want to exercise by this code

% img will be uint8 type
img = imread('syz.bmp');


% imshow: when the parameter img is uint8 type, 
% imshow will default make the range of pixel value as [0-255]
imshow(img);
fprintf('Program paused. Press enter to continue.\n');
pause;

% sometimes we need to process the img by double type,
% so, you may convert the data type to double
dimg = double(img);

% but, right now you will not get the correct display,
% because if the parameter of imshow is double type, if will defaultly 
% take range of [0-1], but now the range is [0-255]
% all the value over 1 is ceilled to 1. (clamped to 1)
% so, the displayed image will be a whole white picture.
imshow(dimg);
fprintf('Program paused. Press enter to continue.\n');
pause;

% how to correctly display double typed image
% way 1: convert double to uint8
imshow(uint8(dimg));
fprintf('Program paused. Press enter to continue.\n');
pause;

% way 2: change the value of double typed image to [0-1]
maxVal = max(max(dimg));
imshow(dimg / maxVal);
fprintf('Program paused. Press enter to continue.\n');
pause;

%% some other occurence, the image maybe in double type, 
% but some normalization operation will lead to negative pixel value
% In order to display the image, we need to add a value that make
% all negative value to positive value

% here i just minus a specific value, in real application,
% a face image maybe normalized by substract mean face.
normImg = dimg - 127; % then normImg will have some negative values

minVal = min(min(normImg));
imshow(uint8(normImg - minVal));
fprintf('Program paused. Press enter to continue.\n');
pause;


%% if you want to save image by imwrite()
% if the image is double type, you need to normalize the value to [0-1]
% if the image is uint8 type, it's ok to save image directly.