看这篇文章的道友想必对高斯滤波已经很熟悉,在此就不进行赘述了,也可以看看参考资料的文章回顾一下。
高斯函数分离特性
二维方式是根据kernel的大小以及sigma大小生成一个 size*size的卷积核,然后再做卷积。计算量是imgWidth * imgHeight * size * size,但如果用两个一维来替代,则计算量是imgWidth * imgHeight * size * 2,计算量大大减少。
代码实现
此处的数据类型ImagePro可以根据自己的需求进行定义。
double *GetGaussianKernel_1D(int arr_size, double sigma)
{
double *array = new double[arr_size];
int center_i = arr_size / 2;
double sum = 0.0f;
double sigma2 = (2.0f*sigma*sigma);
for (int i = 0; i < arr_size; i++)
{
array[i] = exp(-(1.0f)* (((i - center_i)*(i - center_i)) / sigma2));
sum += array[i];
}
//归一化求权值
for (int i = 0; i < arr_size; i++)
{
array[i] /= sum;
cout << array[i] << " ";
}
cout << endl;
return array;
}
void GaussBlur_x(ImagePro* src, ImagePro* dst, int w, int h, int kernelSize, double *gaussianArray)
{
int center = kernelSize / 2;
int ind = 0;
for (int i = 0; i < h; i++)
{
for (int j = 0; j < w; j++)
{
double sum = 0.0;
for (int k = -center; k <= center; k++)
{
if (j + k < 0 || j + k >= w)
continue;
ind = i*w + j + k;
sum += src[ind] * gaussianArray[k + center];
}
// 放入中间结果
ind = ind = i*w + j;
dst[ind] = MAX(MIN(sum, 255), 0);
}
}
}
void GaussBlur_y(ImagePro* src, ImagePro* dst, int w, int h, int kernelSize, double *gaussianArray)
{
int center = kernelSize / 2;
int ind = 0;
for (int i = 0; i < h; i++)
{
for (int j = 0; j < w; j++)
{
double sum = 0.0;
for (int k = -center; k <= center; k++)
{
if (i + k < 0 || i + k >= h)
continue;
ind = (i+k)*w + j;
sum += src[ind] * gaussianArray[k + center];
}
// 放入中间结果
ind = ind = i*w + j;
dst[ind] = MAX(MIN(sum, 255), 0);
}
}
}
Mat FastGaussian(cv::Mat _src, int kernelSize, double sigma)
{
int srcW = _src.cols;
int srcH = _src.rows;
int radius = kernelSize / 2;
//padding
Margin margin;
margin.down = radius;
margin.up = radius;
margin.left = radius;
margin.right = radius;
int w = _src.cols + margin.left + margin.right;
int h = _src.rows + margin.up + margin.down;
Mat pad(h, w, _src.type(), Scalar(0));
ImagePadding((ImagePro*)_src.data, _src.cols, _src.rows, (ImagePro*)pad.data, margin);
_src = pad.clone();
double *gaussianArray = GetGaussianKernel_1D(kernelSize, sigma);
int center = kernelSize / 2;
cv::Mat temp = _src.clone();
cv::Mat dst = _src.clone();
int channels = _src.channels();
// X方向
GaussBlur_x((ImagePro*)_src.data, (ImagePro*)temp.data, _src.cols, _src.rows, kernelSize, gaussianArray);
// Y方向
GaussBlur_y((ImagePro*)temp.data, (ImagePro*)dst.data, _src.cols, _src.rows, kernelSize, gaussianArray);
delete[] gaussianArray;
return dst(Rect(radius, radius, srcW, srcH));
}
参考资料
[1] Opencv之分离高斯滤波
[2] 图像处理之高斯滤波:原理、代码实现和优化加速
[3] 高斯滤波(GaussianFilter)原理及C++实现
[4] 快速高斯滤波、高斯模糊、高斯平滑
[5] 快速高斯滤波(用两个一维代替二维)