C/C++ rotate an BMP image

Question

I'm trying to rotate an BMP image using C/C++ but it's not working.

I've made some functions for read, write and rotation. The read and write functions are working perfectly fine but not rotate from some reason.

EDIT(sin, cos and rotate function)

BMP struct:

struct BMP {
    int width;
    int height;
    unsigned char header[54];
    unsigned char *pixels;
    int size;
};

WRITE:

void writeBMP(string filename, BMP image) {
    string fileName = "Output Files/" + filename;
    FILE *out = fopen(fileName.c_str(), "wb");
    fwrite(image.header, sizeof(unsigned char), 54, out);
    int i;
    unsigned char tmp;
    for (i = 0; i < image.size; i += 3) {
        tmp = image.pixels[i];
        image.pixels[i] = image.pixels[i + 2];
        image.pixels[i + 2] = tmp;
    }
    fwrite(image.pixels, sizeof(unsigned char), image.size, out); // read the rest of the data at once
    fclose(out);
}

READ:

BMP readBMP(string filename) {
    BMP image;
    int i;
    string fileName = "Input Files/" + filename;
    FILE *f = fopen(fileName.c_str(), "rb");
    fread(image.header, sizeof(unsigned char), 54, f); // read the 54-byte header

    // extract image height and width from header
    image.width = *(int *) &image.header[18];
    image.height = *(int *) &image.header[22];

    image.size = 3 * image.width * image.height;
    image.pixels = new unsigned char[image.size]; // allocate 3 bytes per pixel
    fread(image.pixels, sizeof(unsigned char), image.size, f); // read the rest of the data at once
    fclose(f);

    for (i = 0; i < image.size; i += 3) {
        unsigned char tmp = image.pixels[i];
        image.pixels[i] = image.pixels[i + 2];
        image.pixels[i + 2] = tmp;
    }
    return image;
}

ROTATE:

BMP rotate(BMP image, double degree) {
    BMP newImage = image;
    unsigned char *pixels = new unsigned char[image.size];

    double radians = (degree * M_PI) / 180;
    int sinf = (int) sin(radians);
    int cosf = (int) cos(radians);

    double x0 = 0.5 * (image.width - 1);     // point to rotate about
    double y0 = 0.5 * (image.height - 1);     // center of image

    // rotation
    for (int x = 0; x < image.width; x++) {
        for (int y = 0; y < image.height; y++) {
            long double a = x - x0;
            long double b = y - y0;
            int xx = (int) (+a * cosf - b * sinf + x0);
            int yy = (int) (+a * sinf + b * cosf + y0);

            if (xx >= 0 && xx < image.width && yy >= 0 && yy < image.height) {
                pixels[(y * image.height + x) * 3 + 0] = image.pixels[(yy * image.height + xx) * 3 + 0];
                pixels[(y * image.height + x) * 3 + 1] = image.pixels[(yy * image.height + xx) * 3 + 1];
                pixels[(y * image.height + x) * 3 + 2] = image.pixels[(yy * image.height + xx) * 3 + 2];
            }
        }
    }
    newImage.pixels = pixels;
    return newImage;
}

MAIN:

int main() {
    BMP image = readBMP("InImage_2.bmp");
    image = rotate(image,180);
    writeBMP("Output-11.bmp", image);
    return 0;
}

That sin=0.8939966636(in radians) and cos=-0.44807361612(in radians) means this image should be rotated by 90 degree.

Here it's my original image:

And here it's moment my result in:

Could someone help me understand what I'm doing wrong here? I really need this function to work.

I cannot use any third party libraries for this code.

Answer 1

It must handle the rotation in the same pixel format the bmp uses. You are only transforming one byte for each pixel. The pixels look to be wider. This should be easy to fix now that the problem has been identified.

If you need more speed, notice that you have invariants (x and y) that are incremented for each iteration:

for (int y = 0; y < image.height; y++) {
        double a = x - x0;
        double b = y - y0;
        int xx = (int) (+a * cos - b * sin + x0);

Move the b outside of the loop and change the multiplication to addition:

double b = -y0;
for (int y = 0; y < image.height; ++y) {
    int xx = (int) (a * cos - b + x0);
    b += sin;

Notice that the a * cos is a constant for the whole y-loop? Fuse it to the b. Do the same for x0.

double b = a * cos - y0 + x0;
for (int y = 0; y < image.height; ++y) {
    int xx = (int) (- b);
    b += sin;

Notice that the -b costs as well? Negate b.

double b = -(a * cos - y0 + x0);
for (int y = 0; y < image.height; ++y) {
    int xx = (int) b;
    b -= sin;

See what we did there? Next: get rid of the doubles. Use fixed-point. Float-to-integer conversions can be costly. At best, they are useless here.

Last but not least, you are writing into memory vertically. This is very, very bad for the write combine and will dramatically reduce the performance. Consider changing the loop order so that x-loop is the innermost one.

Extra: use tiled memory layout for the image to improve memory locality when reading. Cache will work more efficiently. Not super important here as you only process the image once and tiling would be more expensive than the speed up. But if you want to animate the rotation then tiling should benefit you. Also, with tiling the performance does not fluctuate depending on the rotation angle so the animation will be more consistent (and faster).

EDIT: add explanation how to support more bytes per pixel:

pixels[(y * image.height + x) * 3 + 0] = image.pixels[(yy * image.height + xx) * 3 + 0];
pixels[(y * image.height + x) * 3 + 1] = image.pixels[(yy * image.height + xx) * 3 + 1];
pixels[(y * image.height + x) * 3 + 2] = image.pixels[(yy * image.height + xx) * 3 + 2];

This is starting to be a bit hard to read, but you do see what we are doing there?