Explanation for ddepth parameter in cv2.filter2d() opencv?

Question

import cv2
import numpy as np
from matplotlib import pyplot as plt

img = cv2.imread('logo.png')

kernel = np.ones((5, 5), np.float32) / 25
dst = cv2.filter2D(img, -1, kernel)
plt.subplot(121), plt.imshow(img), plt.title('Original')
plt.xticks([]), plt.yticks([])
plt.subplot(122), plt.imshow(dst), plt.title('Averaging')
plt.xticks([]), plt.yticks([])
plt.show()

I was trying smoothing a picture and i didnt understand the ddepth parameter of cv2.filter2d() where the value is -1. So what does -1 do and also what does ddpeth mean ?

Answer 1

ddepth

uses depth() function which returns the depth of a point transformed by a rigid transform.

and Depth is the number of bits used to represent color in the image it can be 8/24/32 bit for display which can be denoted as (signed char, unsigned short, signed short, int, float, double).

In OpenCV you typically have those types:

8UC3 : 8 bit unsigned and 3 channels => 24 bit per pixel in total.

8UC1 : 8 bit unsigned with a single channel

32S: 32 bit integer type => int

32F: 32 bit floating point => float

64F: 64 bit floating point => double

https://docs.opencv.org/4.x/d4/d86/group__imgproc__filter.html#filter_depths

What is 'Depth' in Image Processing For more information

Answer 2

According to the official doc:

when ddepth=-1, the output image will have the same depth as the source.

And the valid value of ddepth is limited by the following table:

ddepth table

For example:

cv::Mat src(3, 3, CV_8U3);
cv::Mat dst(3, 3, CV_16S3);
cv::Mat dst2(3, 3, CV_16F3);
cv::Mat kernel(3, 3, CV_8U, cv::Scalar(1));
cv::filter2D(src, dst, CV_16S, kernel); // valid
cv::filter2D(src, dst, CV_16F, kernel); // invalid

Answer 3

Basically there are five methods I know to blur images:

1) use gamma Method

2) create your own kernal (kernal: it is nothing but a numpy array of ones of desired shape) and apply it on images

3) use built_in function of OpenCv

blur_img = cv2.blur(Image_src,Kernal_size)

4) gaussian Blur

Guassian_blur_img=cv2.GuassianBlur(img_src,kernel_size,sigma_value)

5) median Blur

Median_blur_img=cv2.medianBlu(img_src,kernel_size_value)

I personally prefer to use Median blur as it smartly remove the noise from your image such only backgroung of image will only get blurred and other features of images is at is in image like corner will be unchanged.

Answer 4

ddepth

ddepth means desired depth of the destination image

It has information about what kinds of data stored in an image, and that can be unsigned char (CV_8U), signed char (CV_8S), unsigned short (CV_16U), and so on...

type

As for type, the type has information combined from 2 values:

image depth + the number of channels.

It can be for example CV_8UC1 (which is equal to CV_8U), CV_8UC2, CV_8UC3, CV_8SC1 (which is equal to CV_8S) etc.

Further Readings

For more discussion, it can be found in the following two articles

• image type vs image depth
• Matrix depth equals 0
• Detailed - Fixed Pixel Types. Limited Use of Templates

Answer 5

You can see in the doc that ddepth stands for "Destination depth", which is the depth of result (destination) image.

If you use -1, the result (destination) image will have the same depth as the input (source) image.