opencv中常用的線性濾波器--boxFilter(),blur(),GaussianBlur()

1. boxFilter()

下面是opencv官方對boxFilter()函數的介紹。如果均衡化（即normalize==ture，這也是預設值），則其本質是均值濾波。

C++:  void  boxFilter (InputArray  src, OutputArray  dst, int  ddepth, Size  ksize, Point  anchor=Point(-1,-1), bool normalize=true, int  borderType=BORDER_DEFAULT  )

Python:   cv2. boxFilter (src, ddepth, ksize [, dst [, anchor [, normalize [, borderType ] ] ] ] ) → dst

Parameters:

src – input image. dst – output image of the same size and type as src. ddepth – the output image depth (-1 to use src.depth()). ksize – blurring kernel size. anchor – anchor point; default value Point(-1,-1) means that the anchor is at the kernel center. normalize – flag, specifying whether the kernel is normalized by its area or not. borderType – border mode used to extrapolate pixels outside of the image.

The function smoothes an image using the kernel:

where

2. blur()

調用blur()等效于調用将normalize=true的boxFilter().

Blurs an image using the normalized box filter.

C++:  void  blur (InputArray  src, OutputArray  dst, Size  ksize, Point  anchor=Point(-1,-1), int borderType=BORDER_DEFAULT  )

Python:   cv2. blur (src, ksize [, dst [, anchor [, borderType ] ] ] ) → dst

Parameters:

src – input image; it can have any number of channels, which are processed independently, but the depth should be CV_8U, CV_16U, CV_16S, CV_32F or CV_64F. dst – output image of the same size and type as src. ksize – blurring kernel size. anchor – anchor point; default value Point(-1,-1) means that the anchor is at the kernel center. borderType – border mode used to extrapolate pixels outside of the image.

The function smoothes an image using the kernel:

The call blur(src, dst, ksize, anchor, borderType) is equivalent to boxFilter(src, dst, src.type(), anchor,true, borderType) .

3.GaussianBlur()

高斯濾波可以消除高斯噪聲，廣泛應用于圖像處理的減噪過程。需要注意的是opencv中的GaussianBlur()是高斯低通濾波器，用來模糊減噪，是以叫高斯模糊。

整數模闆用的比較多，常見的3x3或者5x5的整數模闆如下。更多高斯濾波的講解可以參考下面這篇部落格http://blog.csdn.net/yansmile1/article/details/46275791

4. 應用執行個體

分别調用上面提到的三個函數對一副圖像進行模糊操作，選取的kernel size為5x5。代碼如下

#include<opencv.hpp>
#include<iostream>

int main(void)
{
	cv::Mat src = cv::imread("d:/Opencv Picture/Lena.jpg", CV_LOAD_IMAGE_GRAYSCALE);

	if (!src.data)
	{
		std::cout << "image read error!!!\n";
	}
	cv::imshow("src",src);

	//call boxFilter()
	cv::Mat boxFilterDst;
	cv::boxFilter(src, boxFilterDst, -1, cv::Size(5, 5));
	cv::imshow("boxFilterResult", boxFilterDst);

	//call blur()
	cv::Mat blurDst;
	cv::blur(src, blurDst, cv::Size(5, 5));
	cv::imshow("blurResult", blurDst);

	//call GaussianBlur()
	cv::Mat GaussianDst;
	cv::GaussianBlur(src, blurDst, cv::Size(5, 5),0.8,0.8);
	cv::imshow("GaussianBlurResult", blurDst);

	cvWaitKey(0);
	
	return 0;
}
           

運作結果如下，明顯在相同的kernel size的情況下，GaussianBlur()的結果對原圖的失真比較少（當然還和sigmaX & sigmaY有關），而boxFilter & blur()得到的結果是相同的，因為都是均值濾波。