scale, resize bmp structure, programming c

Question

my task is to scale bmp picture with the given factor:

stream: base64 bmp image code:

Qk1CAAAAAAAAADYAAAAoAAAABAAAAAEAAAABABgAAAAAAAwAAAAjLgAAIy4AAAAAAAAAAAAA/////wAAAP8AAAAA

scale.c

struct bmp_image* scale(const struct bmp_image* image, float factor){
    if(image == NULL || factor <= 0){
        return NULL;
    }

    .........
    int width = bmp->header->width;
    int height = bmp->header->height;
    int widthNew = (int)round((float)width * factor);
    int heightNew = (int)round((float)height * factor);


    for (uint32_t y = 0; y < heightNew; y++){
        for(uint32_t z = 0; z< widthNew; z++){
            bmp->data[(y*widthNew)+z].red=image->data[(int)(floor((float)y/(float)factor)*width+floor((float)z/(float)factor))].red;
            bmp->data[(y*widthNew)+z].green=image->data[(int)(floor((float)y/(float)factor)*width+floor((float)z/(float)factor))].green;
            bmp->data[(y*widthNew)+z].blue=image->data[(int)(floor((float)y/(float)factor)*width+floor((float)z/(float)factor))].blue;
        }
    }

    return bmp;     
}

main.c

FILE *output_p8 = fopen("scale.bmp", "w");
    struct bmp_image* newimage8 = NULL;
    newimage8 = scale(image, 1.025656);  
    free_bmp_image(image);    
    write_bmp(output_p8,newimage8);
    free_bmp_image(newimage8);    
    fclose(output_p8);

the data output what I got after that:

ffffff ffffff ff0000 00ff00

the data what I should have:

ffffff ff0000 00ff00 000000

any suggest?

Answer 1

When you calculate the new dimensions, you perform a floating-point calculation and then round to the nearest integer:

int widthNew = round(width * factor);

(You don't need the casts: factor is a float and the result of a multiplication of an integer with a float is also a float; widthNew is an integer, so the float result must be converted to an integer.)

Because you round, your effective scaling factor (let's call it factor_eff) may differ from the nominal factor:

float factor_eff = (float) widthNew / width;

In your case:

factor == 1.025656;
width == 4;
widthNew == round(1.025656 * 4) == round(4.102624) == 4;
factor_eff = 4.0 / 4 == 1.0;

But you do your index calculations with the old factor:

int zNew = floor(z / factor);

(Again, you don't need the cast.) For your four z indces, this will yield:

z == 0;    zNew == floor(0 / 1.025656) == floor(0.0)       == 0;   // ok
z == 1;    zNew == floor(1 / 1.025656) ~~ floor(0.9749858) == 0;   // ! should be 1
z == 2;    zNew == floor(2 / 1.025656) ~~ floor(1.9499715) == 1;   // ! should be 2
z == 3;    zNew == floor(3 / 1.025656) ~~ floor(2.9249573) == 2;   // ! should be 3

You could fix the floating-point factor, but I think it is better (and faster) to calculate your new indices with integer arithmetic:

int zNew = z * width / widthNew;
int yNew = y * height / heightNew;

Make sure you do the multiplication first. This method should be safe to use for dimensions up to abour 46,000.