C中的双三次插值

Question

我尝试处理C语言中的双三次插值图像插值。因此，我编写了这个小脚本。

1. "resize_image"函数：

    void resize_image(PPMImage *source_image, PPMImage *destination_image, float scale) {

        uint8_t sample[3];
        int y, x;

        destination_image->x = (long)((float)(source_image->x)*scale);
        destination_image->y = (long)((float)(source_image->y)*scale);

        for (y = 0; y < destination_image->y; y++) {

            float v = (float)y / (float)(destination_image->y - 1);

            for (x = 0; x < destination_image->x; ++x) {

                float u = (float)x / (float)(destination_image->x - 1);
                sample_bicubic(source_image, u, v, sample);

                destination_image->data[x+((destination_image->y)*y)].red   = sample[0];
                destination_image->data[x+((destination_image->y)*y)].green = sample[1];  
                destination_image->data[x+((destination_image->y)*y)].blue  = sample[2];  
            }
        }
    }

2. “sample_bicubic”函数

（涉及IT技术相关内容）

    void sample_bicubic(PPMImage *source_image, float u, float v, uint8_t sample[]) {

        float x = (u * source_image->x)-0.5;
        int xint = (int)x;
        float xfract = x-floor(x);

        float y = (v * source_image->y) - 0.5;
        int yint = (int)y;
        float yfract = y - floor(y);

        int i;

        uint8_t p00[3];
        uint8_t p10[3];
        uint8_t p20[3];
        uint8_t p30[3];

        uint8_t p01[3];
        uint8_t p11[3];
        uint8_t p21[3];
        uint8_t p31[3];

        uint8_t p02[3];
        uint8_t p12[3];
        uint8_t p22[3];
        uint8_t p32[3];

        uint8_t p03[3];
        uint8_t p13[3];
        uint8_t p23[3];
        uint8_t p33[3];

        // 1st row
        get_pixel_clamped(source_image, xint - 1, yint - 1, p00);   
        get_pixel_clamped(source_image, xint + 0, yint - 1, p10);
        get_pixel_clamped(source_image, xint + 1, yint - 1, p20);
        get_pixel_clamped(source_image, xint + 2, yint - 1, p30);

        // 2nd row
        get_pixel_clamped(source_image, xint - 1, yint + 0, p01);
        get_pixel_clamped(source_image, xint + 0, yint + 0, p11);
        get_pixel_clamped(source_image, xint + 1, yint + 0, p21);
        get_pixel_clamped(source_image, xint + 2, yint + 0, p31);

        // 3rd row
        get_pixel_clamped(source_image, xint - 1, yint + 1, p02);
        get_pixel_clamped(source_image, xint + 0, yint + 1, p12);
        get_pixel_clamped(source_image, xint + 1, yint + 1, p22);
        get_pixel_clamped(source_image, xint + 2, yint + 1, p32);

        // 4th row
        get_pixel_clamped(source_image, xint - 1, yint + 2, p03);
        get_pixel_clamped(source_image, xint + 0, yint + 2, p13);
        get_pixel_clamped(source_image, xint + 1, yint + 2, p23);
        get_pixel_clamped(source_image, xint + 2, yint + 2, p33);

        // interpolate bi-cubically!
        for (i = 0; i < 3; i++) {

            float col0 = cubic_hermite(p00[i], p10[i], p20[i], p30[i], xfract);
            float col1 = cubic_hermite(p01[i], p11[i], p21[i], p31[i], xfract);
            float col2 = cubic_hermite(p02[i], p12[i], p22[i], p32[i], xfract);
            float col3 = cubic_hermite(p03[i], p13[i], p23[i], p33[i], xfract);

            float value = cubic_hermite(col0, col1, col2, col3, yfract);

            CLAMP(value, 0.0f, 255.0f);

            sample[i] = (uint8_t)value;

            printf("sample[%d]=%d\n",i,sample[i]);      

        }
    }

3. 插值助手

    float cubic_hermite(float A, float B, float C, float D, float t) {

        float a = -A / 2.0f + (3.0f*B) / 2.0f - (3.0f*C) / 2.0f + D / 2.0f;
        float b = A - (5.0f*B) / 2.0f + 2.0f*C - D / 2.0f;
        float c = -A / 2.0f + C / 2.0f;
        float d = B;

        return a*t*t*t + b*t*t + c*t + d;
    }

    void get_pixel_clamped(PPMImage *source_image, int x, int y, uint8_t temp[])  {

        CLAMP(x, 0, source_image->x - 1);
        CLAMP(y, 0, source_image->y - 1);

        temp[0] = source_image->data[x+(W*y)].red;
        temp[1] = source_image->data[x+(W*y)].green;
        temp[2] = source_image->data[x+(W*y)].blue;
    }

我已上传了完整的代码和所有相关内容在这里。

执行此代码时没有语法错误。

但输出图像让我感到困惑。

输入图像（21x20像素）：

此输入图像被放大了2倍（42x40像素）：

插值似乎在某些点上运行良好，但图像看起来像像素被移动了。

有人能告诉我我做错了什么吗？ 此脚本是在以下帮助下制作的： http://blog.demofox.org/2015/08/15/resizing-images-with-bicubic-interpolation/

谢谢大家！

（请不要考虑此代码的效率...我知道它很糟糕）

- Felix K.

---

哈哈... #include <windows.h> // 用于位图头。抱歉，非Windows用户！ - dAm2K

Answer 1

从你的resize_image()函数：

destination_image->data[x+((destination_image->y)*y)].red   = sample[0];

那可能应该是

destination_image->data[x+((destination_image->x)*y)].red   = sample[0];

在调试此类问题时有用的是，使用一些在实际数据中不存在的“魔法颜色”来初始化目标图像（例如，一些可怕的粉色:-)）。然后，你可能会注意到一些目标像素在调用resize_image()函数后仍具有该颜色。这提示了问题所在。

- ThorngardSO

Answer 2

我测试了你的代码，发现还有一个问题。

从你的init_destination_image()函数中：

img->data = (PPMPixel*)malloc(W * H * (int)scale * sizeof(PPMPixel));

应该是：

应该是：

img->data = (PPMPixel*)malloc(W * H * (int)scale * (int)scale * sizeof(PPMPixel));

例如，W 和 H 都被缩放。

- Joseph Pena