CN103150710A

CN103150710A - Rapid mean filtering method for image

Info

Publication number: CN103150710A
Application number: CN2013100965429A
Authority: CN
Inventors: 夏永泉; 支俊; 黄敏; 甘勇; 张娜
Original assignee: Zhengzhou University of Light Industry
Current assignee: Zhengzhou University of Light Industry
Priority date: 2013-03-25
Filing date: 2013-03-25
Publication date: 2013-06-12

Abstract

The invention relates to a rapid mean filtering method for an image. At first, neighboring mean values of pixels f(2n, 2n) are calculated; next, the neighboring mean values of other pixels in a row or a column are obtained through the utilization of an intermediate result of the neighboring mean values of the pixels f(2n, 2n), so that the purpose of saving calculation time is achieved; and moreover, the sizes of neighboring windows are irrelevant. The experiment result shows that the rapid mean filtering method for the image is high in speed and high in robustness, and enables the complexity of mean value calculation to be effectively decreased, thereby having great values and significances in realizing rapid mean value filtering during the image processing.

Description

A kind of Quick and equal filtering method for image

Technical field

The invention belongs to image processing field, be specifically related to rapid image mean filter method.

Background technology

In the acquisition process of image, be subjected to the impact of factors, can cause the degeneration of picture quality.Simultaneously, in the transmitting procedure of image, also can produce some noises.The purpose of figure image intensifying is exactly to take a series of technology to improve the visual effect of image, and mean filter is a kind of image enhancement technique of directly carrying out on the image space territory, is a kind of basic skills of eliminating noise in the level and smooth preconditioning technique of image filtering.Be a kind of Image Denoising Technology commonly used in image is processed, but computation complexity is high, affects system real time.

Summary of the invention

The technical problem to be solved in the present invention reduces the computation complexity of mean filter, and a kind of Quick and equal filtering method for image is provided.

Technical scheme of the present invention is: a kind of Quick and equal filtering method for image, analyze the computation process of current pixel and neighbor average, take full advantage of the results of intermediate calculations of current pixel, draw the mutual relationship of mean value computation, eliminate redundant computation, its step is as follows:

(1) impose a condition: suppose that the filter window size is (2n+1) * (2n+1), obtains the pixel data of image in computing machine;

(2) calculating f(2n, 2n) neighboring mean value N(2n, the 2n of pixel), wherein N( x,y) expression with ( x,y) centered by size be

The set of neighborhood window pixel;

(3) neighborhood of pixels of calculated level direction, with ( x,y) neighbor ( x+1 , y) neighboring mean value

Be expressed as:

Wherein Col( X-n, y)----------expression with ( x,y) centered by the 1st row row pixel of neighborhood window, Col(( X+1)+n, y)---the expression with ( X+1, y) centered by the 2n+1 row row pixel of neighborhood window,

----------neighboring mean value;

(4) calculate the neighborhood of pixels average of vertical direction, along x column count neighboring mean value the time, With

In comprised the part of double counting, keep away redundant computation by following strategy:

Wherein Row( X, y-n)--------expression with ( x,y) centered by the 1st row pixel of neighborhood window, Row( X, (y+1)+n)--the expression with ( X, y+1) centered by the capable pixel of 2n+1 of neighborhood window;

(5) according to f(2n, 2n) neighboring mean value, analyze the double counting part, by the neighboring mean value of each pixel of pixel calculated level direction.

As long as calculate the neighboring mean value of a pixel of the first row first row, the neighboring mean value that remains other each pixel only relates to the simple computation of the two individual pixels of row 2 (2n+1) like this, originally needs to calculate (2n+1) ²Thereby the inferior purpose that reaches the algorithm acceleration.

The invention has the beneficial effects as follows: the present invention has adopted respectively common mean filter method and the fast method in literary composition to compare, and result of the present invention shows, the method in literary composition is not adapted the effect of filtering, but the travelling speed of algorithm improves greatly.As shown in table 1, concerning 3 * 3 neighborhood windows, commonsense method 63ms, and method 49ms in literary composition, travelling speed has improved 22.22%, and when adopting 7 * 7 window, speed has improved 69.5%, and the field window is larger, and the time of saving is more.

Description of drawings

Fig. 1 is source images of the present invention;

Fig. 2 is the image that the present invention has added salt-pepper noise;

Fig. 3 is the result after employing 3 * 3 mean filters of the embodiment of the present invention 2;

Fig. 4 is the result after employing 7 * 7 mean filters of the embodiment of the present invention 3.

Embodiment

Embodiment 1

A kind of Quick and equal filtering method for image, its step is as follows:

(2) calculating f(2n, 2n) neighboring mean value N(2n, the 2n of pixel), wherein N( x,y) expression with ( x,y) centered by size be The set of neighborhood window pixel;

Be expressed as:

Wherein Col( X-n, y) expression with ( x,y) centered by the 1st row row pixel of neighborhood window, Col(( X+1)+n, y)---the expression with ( X+1, y) centered by the 2n+1 row row pixel of neighborhood window,

----------neighboring mean value;

(4) calculate the neighborhood of pixels average of vertical direction, along x column count neighboring mean value the time,

With

Wherein Row( X, y-n) expression with ( x,y) centered by the 1st row pixel of neighborhood window, Row( X, (y+1)+n) expression with ( X, y+1) centered by the capable pixel of 2n+1 of neighborhood window;

Embodiment 2

A kind of Quick and equal filtering method for image, its step is as follows:

(1) the input size is the lena image of 256 * 256 sizes;

(2) setting the neighborhood window size is 3 * 3;

(3) image is added salt-pepper noise;

(4) calculate f(2,2) neighboring mean value of pixel

(5) according to f(2, the 2) neighboring mean value of pixel calculates the neighboring mean value of each pixel of horizontal direction of the 2nd row;

(6) according to f(2, the 2) neighboring mean value of pixel calculates the neighboring mean value of each pixel of vertical direction of the 2nd row.

Experimental data is as shown in table 1.

Table 1 experimental data

Embodiment 3

A kind of Quick and equal filtering method for image, its step is as follows: setting the neighborhood window size is 7 * 7, method and step only need the mean value computation window size is changed to 7 * 7 with embodiment 2.

Claims

1. a Quick and equal filtering method that is used for image, is characterized in that, its step is as follows:

Figure 2013100965429100001DEST_PATH_IMAGE002

The set of neighborhood window pixel;

(3) neighborhood of pixels of calculated level direction, with ( x,y) neighbor ( x+1 , y) neighboring mean value Be expressed as:

Figure 2013100965429100001DEST_PATH_IMAGE008

----------neighboring mean value;

Figure 2013100965429100001DEST_PATH_IMAGE012

Figure 2013100965429100001DEST_PATH_IMAGE014