Saving images of different depths in opencv

What is image depth

A lot of people start to learn OpenCV after , Touch the front few API Include imwrite function , And soon learned how to save Mat Object is image , The general code is as follows ：

imwrite("D:/result.png ", dst);

among dst yes Mat object .

The image saved in this way is each channel by default 8 Bit byte image , common RGB The image is an image with a depth of 24, This can be done through windows View the image properties to get , The screenshot is as follows :

If each channel occupies 16 A word of ,RGB The image depth becomes 48, If each channel occupies 32 A word of , The depth becomes 96, Obviously, the deeper the image , There will also be more image files , The more memory will be consumed when loading , therefore OpenCV The default read / write image in is each channel 8 position ( Single byte ) Images .

To relive imwrite function

Suppose we want to save the image as 16 Bits or 32 Bit floating point number image , What should we do , Before we begin, let's review again OpenCV Save image in API function -imwrite

bool cv::imwrite(
   const String &  filename,
   InputArray img,
   const std::vector< int > &  params = std::vector< int >()
)
Filename  The parameter is the path to save the declared file 
Img The parameter represents the... To be saved Mat Image object 
Params  Represents the options when saving an image ,

These options include PNG/JPG/WEBP/TIFF The compression quality 、 Format selection, etc , It can be divided into the following four categories

• ImwriteEXRTypeFlags

• ImwriteFlags

• ImwritePAMFlags

• ImwritePNGFlags,

I wrote an article before about how to use these options to compress the image properly when maintaining , Here's about Params The use of parameters will not be repeated , Mainly based on key-value Method added to vector Just go to the middle .

imwrite The image type support of the function when saving to different depth formats is described as follows ：

• 8 Bitwise image (CV_8U), Support png/jpg/bmp/webp And other common image formats

• 16 Bitwise image (CV_16U), Support png/jpeg2000/TIFF Format

• 32 Bitwise image (CV_32F), Support PFM/TIFF/OpenEXR/TIFF/HDR

Before saving to the specified format , Can pass convertTo perhaps cvtCOLOR After image type or channel conversion , Call again imwrite Preservation .

Various depths of preservation

16 Bit image saving

After the transformation , If you save directly , The code is as follows ：

//  Load image 
Mat src = imread("D:/flower.png", IMREAD_UNCHANGED);
printf("depth %d \n", src.depth());

//  Convert to 16 Bit image 
Mat dst;
src.convertTo(dst, CV_16U);
imshow("flower16", dst);
imwrite("D:/flower-16.png", dst);

After the transformation , Normalize and save , The code is as follows ：

//  Load image 
Mat src = imread("D:/flower.png", IMREAD_UNCHANGED);
printf("depth %d \n", src.depth());

//  Convert to 16 Bit image 
Mat dst;
src.convertTo(dst, CV_16U);

//  Normalize and save 
normalize(dst, dst, 0, 256 * 256, NORM_MINMAX);
imwrite("D:/flower-16.png", dst);
imshow("flower-16", dst);

The comparison of the two effects is as follows ：

32 Bit image saving

//  Load image 
Mat src = imread("D:/flower.png", IMREAD_UNCHANGED);
printf("depth %d \n", src.depth());

//  Convert to 16 Bit image 
Mat dst;
src.convertTo(dst, CV_32F);

//  Normalize and save 
normalize(dst, dst, 0, 1.0, NORM_MINMAX);
imwrite("D:/flower-32.png", dst);
imshow("flower-32", dst);

For the above images with different depths , It must be read correctly in the following way

Mat src = imread("D:/flower.png", IMREAD_UNCHANGED);

among IMREAD_UNCHANGED Indicates that no changes will be made to the original image .

link