CN105260994A - Aerial insulator image de-noising method - Google Patents

Abstract

The invention relates to an aerial insulator image de-noising method. A basic image processing method is comprehensively used, aerial image noise features are analyzed, and an aerial image de-noising method capable of removing mixed pulse noise and Gaussian noise is designed. The latest and typical research result in the current image processing field is introduced to power system helicopter line inspection, aerial image noise features are used, and the noise image is processed in a targeted mode; and key theory and application basis is laid for transmission line defect detection and diagnosis based on helicopter aerial photographing or robot inspection, and the application prospect is good.

Description

A kind of denoising method of insulator image of taking photo by plane

[technical field]

The present invention relates to a kind of denoising method of insulator image of taking photo by plane, belong to electric system helicopter routing inspection or robot patrols and examines technical field.

[background technology]

For a long time, due to the restriction by factors such as geographical environments, power transmission line inspection work is mainly observed by artificial visually examine, and this traditional working method can only find the defect bottom some ground, shaft tower, and some fairly simple defects of easily seeing such as glass insulator self-destruction, work efficiency is lower.Along with the appearance of high voltage, high-power, long distance transmission line, the geographical environment that transmission line of electricity passes through is increasingly sophisticated, brings a lot of difficulty to line data-logging, and mode automatic detecting transmission line of electricity such as application helicopter or robot etc. has become the trend of electric power development.

Helicopter or robot patrol and examine general record infrared image and visible images, and corresponding thermal imagery is maked an inspection tour and visible ray tour respectively.Have that scope is large, vision is wide, efficiency is high from air observation, not by advantages such as regional impacts, but by Atmospheric Flow, sunlight irradiates, and the impact of the shooting environmental such as fuselage shaking, be usually mixed into Gaussian noise and impulsive noise in shooting image.These noises not only can hinder people to the understanding of image, and cause larger impact to follow-up image processing work, therefore, remove Aerial Images noise and just seem particularly important.

At present, for Aerial Images Denoising Problems, scholars propose many methods, as wavelet transformation, PDE method, coefficient table method etc.But said method all exists many defects, be not well positioned to meet existing needs.

[summary of the invention]

The object of the invention is to: according to this fact being often mixed with Gaussian noise and impulsive noise in Aerial Images; propose a kind of denoising method of insulator image of taking photo by plane herein; because in insulator picture noise of taking photo by plane, impulsive noise and Gaussian noise are common type; the classification to unlike signal can be realized by pixel cohort; utilize medium filtering and Wiener filtering; by the in addition cancellation of the Gaussian noise of having classified and impulsive noise, retain original image to greatest extent simultaneously.

For achieving the above object, the technical solution used in the present invention is: a kind of denoising method of insulator image of taking photo by plane, comprises the following steps:

(1), the pixel cohort grouping of noise image

For a given pixel p, its pixel cohort is defined as a set similar to it in its field; According to Fisher discriminant, find out the cohort of pixel, self-adaptation obtains the number of cohort pixel;

If the size of image is , center pixel is ,

The objective function of Fisher discriminant is:

Wherein, , , ,

By calculating each t value , when time maximum, corresponding t value is cohort number p, is the cohort pixel of window center pixel;

(2), the rim detection of original image

Utilize sobel edge detection method, obtain the edge basis for estimation of original image, in the image after rim detection, if certain some pixel value is 1, then represent that this point is marginal point in original image; If this pixel value is 0, then represent that this point is non-edge point in original image;

(3), the noise of noise image judges

By cohort number p, the noise of image zones of different is judged, judges as follows:

If A , then can judge that the cohort pixel of the center pixel of window only has 1, i.e. itself, now can be judged as impulsive noise;

If B , then can illustrate that the cohort of center pixel comprises all pixels in window, no edge pixel when now judging this pixel, if not, then can be judged as Gaussian noise, if so, then not deal with;

(4), the process of impulsive noise

If pending pixel is in 8 directions on neighbor be: , then intermediate value y is defined as:

Then y value is that the medium filtering of this pixel is newly worth;

(5), the process of Gaussian noise

Gaussian noise utilizes Wiener filtering algorithm to process, and the step of Wiener filtering is as follows:

A, the local mean value estimating pixel and variance

Wherein, being the size of each pixel in image is field;

B, utilize S filter estimated image signal

Wherein, the noise variance of entire image, if without knowledge of noise covariance, then get the mean value of the local variance of all pixels.

After adopting said method, beneficial effect of the present invention is: the present invention can realize the classification to unlike signal by pixel cohort, utilize medium filtering and Wiener filtering, by the cancellation in addition of the Gaussian noise of having classified and impulsive noise, retain original image to greatest extent simultaneously, for the transmission line of electricity defects detection and diagnosis of patrolling and examining video based on helicopter aerial photography or robot establish critical theory and application foundation, there is extraordinary application prospect.

[accompanying drawing explanation]

Accompanying drawing described herein is used to provide a further understanding of the present invention, forms a application's part, but does not form inappropriate limitation of the present invention, in the accompanying drawings:

The original image that Fig. 1 takes photo by plane;

Fig. 2 original image sobel edge detection results image;

The noise image that Fig. 3 takes photo by plane;

Result images after the denoising of Fig. 4 noise image.

[embodiment]

Describe the present invention in detail below in conjunction with accompanying drawing and specific embodiment, illustrative examples wherein and explanation are only used for explaining the present invention, but not as a limitation of the invention.

The feature of insulator image of taking photo by plane comprises as follows:

1, insulator shows as linear string in the picture in most cases;

2, insulator shows as elliptical shape in the picture in most cases;

3, to take photo by plane normal Noise in insulator image;

4, impulsive noise and the Gaussian noise in insulator picture noise of taking photo by plane is common type.

A kind of denoising method of taking photo by plane insulator image proposed by the invention develops according to the feature of above-mentioned Aerial Images.As Figure 1-4, a kind of denoising method of insulator image of taking photo by plane, comprises the following steps:

(1), the pixel cohort grouping of noise image

For a given pixel p, its pixel cohort is defined as a set similar to it in its field; According to Fisher discriminant, find out the cohort of pixel, self-adaptation obtains the number of cohort pixel;

If the size of image is , center pixel is ,

The objective function of Fisher discriminant is:

Wherein, , , ,

By calculating each t value , when time maximum, corresponding t value is cohort number p, is the cohort pixel of window center pixel;

(2), the rim detection of original image

Utilize sobel edge detection method, obtain the edge basis for estimation of original image, in the image after rim detection, if certain some pixel value is 1, then represent that this point is marginal point in original image; If this pixel value is 0, then represent that this point is non-edge point in original image;

(3), the noise of noise image judges

By cohort number p, the noise of image zones of different is judged, judges as follows:

If A , then can judge that the cohort pixel of the center pixel of window only has 1, i.e. itself, now can be judged as impulsive noise;

If B , then can illustrate that the cohort of center pixel comprises all pixels in window, no edge pixel when now judging this pixel, if not, then can be judged as Gaussian noise, if so, then not deal with;

(4), the process of impulsive noise

If pending pixel is in 8 directions on neighbor be: , then intermediate value y is defined as:

Then y value is that the medium filtering of this pixel is newly worth;

(5), the process of Gaussian noise

Gaussian noise utilizes Wiener filtering algorithm to process, and the step of Wiener filtering is as follows:

A, the local mean value estimating pixel and variance

Wherein, being the size of each pixel in image is field;

B, utilize S filter estimated image signal

Wherein, the noise variance of entire image, if without knowledge of noise covariance, then get the mean value of the local variance of all pixels.

This invention, based on MATLAB development platform, reads original image, with the edge detection operator of sobel as edge edge indicator function, obtains edge detection results figure; Read noise image, utilize said method of the present invention, noise image is carried out denoising.

The above is only better embodiment of the present invention, therefore all equivalences done according to structure, feature and the principle described in patent claim of the present invention change or modify, and are included in patent claim of the present invention.

Claims (1)

1. to take photo by plane the denoising method of insulator image, it is characterized in that: comprise the following steps:

(1), the pixel cohort grouping of noise image

For a given pixel p, its pixel cohort is defined as a set similar to it in its field; According to Fisher discriminant, find out the cohort of pixel, self-adaptation obtains the number of cohort pixel;

If the size of image is , center pixel is ,

The objective function of Fisher discriminant is:

Wherein, , , ,

By calculating each t value , when time maximum, corresponding t value is cohort number p, is the cohort pixel of window center pixel;

(2), the rim detection of original image

Utilize sobel edge detection method, obtain the edge basis for estimation of original image, in the image after rim detection, if certain some pixel value is 1, then represent that this point is marginal point in original image; If this pixel value is 0, then represent that this point is non-edge point in original image;

(3), the noise of noise image judges

By cohort number p, the noise of image zones of different is judged, judges as follows:

If A , then can judge that the cohort pixel of the center pixel of window only has 1, i.e. itself, now can be judged as impulsive noise;

If B , then can illustrate that the cohort of center pixel comprises all pixels in window, no edge pixel when now judging this pixel, if not, then can be judged as Gaussian noise, if so, then not deal with;

(4), the process of impulsive noise

If pending pixel is in 8 directions on neighbor be: , then intermediate value y is defined as:

Then y value is that the medium filtering of this pixel is newly worth;

(5), the process of Gaussian noise

Gaussian noise utilizes Wiener filtering algorithm to process, and the step of Wiener filtering is as follows:

A, the local mean value estimating pixel and variance

Wherein, being the size of each pixel in image is field;

B, utilize S filter estimated image signal

Wherein, the noise variance of entire image, if without knowledge of noise covariance, then get the mean value of the local variance of all pixels.