How to get images from raster bands using OpenCV? - opencv

Question

I have different raster bands and I need to get a proper image from these bands using OpenCV. How can it be done?

Answer 1

As an alternative to my other answer, you could put a small ASCII header on your raw files to make them into PGM files which OpenCV can read natively without any extra libraries. The PGM format is part of the NetPBM suite described here on Wikipedia.
So, let's say each of your rasters is a single channel, 8-bit image with dimensions 100 px wide by 256 px tall. Make each band into a PGM image in the Terminal:
{ printf "P5\n100 256\n255\n"; cat band1.dat; } > band1.pgm
{ printf "P5\n100 256\n255\n"; cat band2.dat; } > band2.pgm
{ printf "P5\n100 256\n255\n"; cat band3.dat; } > band3.pgm
{ printf "P5\n100 256\n255\n"; cat band4.dat; } > band4.pgm
You now have 4 greyscale images you can view with GIMP or feh and, more importantly which OpenCV can read. So your code becomes:
Mat b1 = imread("band1.pgm", IMREAD_UNCHANGED);
Mat b2 = imread("band2.pgm", IMREAD_UNCHANGED);
Mat b3 = imread("band3.pgm", IMREAD_UNCHANGED);
Mat b4 = imread("band4.pgm", IMREAD_UNCHANGED);
// Now merge
auto channels = std::vector<cv::Mat>{b1, b2, b3, b4};
cv::Mat FourBandBoy;
cv::merge(channels, FourBandBoy);
As you haven't provided any sample raster images, I made 4 images, each 100x256, for a quick demonstration. Here they are:
-rw-r--r-- 1 mark staff 25600 21 Mar 10:44 band1.dat
-rw-r--r-- 1 mark staff 25600 21 Mar 10:44 band2.dat
-rw-r--r-- 1 mark staff 25600 21 Mar 10:44 band3.dat
-rw-r--r-- 1 mark staff 25600 21 Mar 10:44 band4.dat
Hopefully, you can see from the sizes that they are 100x256, single channel and 8-bit.
I the converted them to PGM per the original instructions:
{ printf "P5\n100 256\n255\n"; cat band1.dat; } > band1.pgm
{ printf "P5\n100 256\n255\n"; cat band2.dat; } > band2.pgm
{ printf "P5\n100 256\n255\n"; cat band3.dat; } > band3.pgm
{ printf "P5\n100 256\n255\n"; cat band4.dat; } > band4.pgm
which gives:
-rw-r--r-- 1 mark staff 25615 21 Mar 10:44 band1.pgm
-rw-r--r-- 1 mark staff 25615 21 Mar 10:44 band2.pgm
-rw-r--r-- 1 mark staff 25615 21 Mar 10:44 band3.pgm
-rw-r--r-- 1 mark staff 25615 21 Mar 10:44 band4.pgm
So, you can see the PGM header amounts to 15 bytes. The images look like this now:
I slightly modified the code:
#include <iostream>
#include <opencv2/opencv.hpp>
#include <cstdio>
using namespace cv;
using namespace std;
int
main(int argc,char*argv[])
{
Mat b1 = imread("band1.pgm", IMREAD_UNCHANGED);
Mat b2 = imread("band2.pgm", IMREAD_UNCHANGED);
Mat b3 = imread("band3.pgm", IMREAD_UNCHANGED);
Mat b4 = imread("band4.pgm", IMREAD_UNCHANGED);
// Now merge
auto channels = std::vector<cv::Mat>{b1,b2,b3,b4};
cv::Mat BigBoy;
cv::merge(channels, BigBoy);
// Save
cv::imwrite("result.png",BigBoy);
}
And the result...
Hopefully you can see that the resulting image:
is blue where there was a lot of white in band1,
is green where there was a lot of white in band2,
is red where band3 was white
is transparent where band 4 was black

Answer 2

I don't feel like writing all the code and testing it like I ordinarily might so I'll just give you an outline of one way to do it.
Assume your images are WIDTHxHEIGHT in dimension and made of unsigned char data written consecutively without padding into files called band1.dat, band2.dat etc.
// Step 1: Allocate a buffer: raw = WIDTHxHEIGHT bytes
// Step 2: Load one band from disk into raw buffer
open band1.dat, read WIDTH*HEIGHT bytes into raw, close band1.dat
// Make a single band Mat from that data
cv::Mat band1(HEIGHT,WIDTH,CV_8UC1);
// Copy the raw data into it so we can re-use buffer
std::memcpy(band1.data,raw,WIDTH*HEIGHT);
// Repeat step 2 for bands 2-4 so you have 4 separate, single channel Mats
// Now merge the 4 individual channels into 4-band bad boy
auto channels = std::vector<cv::Mat>{band1, band2, band3, band4};
cv::Mat FourBandBoy;
cv::merge(channels, FourBandBoy);