CN114885107A - Method for correcting noise of image fixed mode - Google Patents

Abstract

The invention relates to a method for correcting noise of an image fixed mode, which comprises the following steps of 1, acquiring image data to obtain Fmean and Fstd; 2. obtaining an average value Nave; 3. solving the mean value NAve of the position P data; 4. solving a data mean Ivar of the position P; 5. setting a decision threshold Flim =2 NAve-Nave; 6. correcting the image; 7. and repeating the traversal, and repeating the steps 2 to 6 in sequence until the traversal correction of the image column is finished. The method for correcting the noise of the image fixed mode is based on a method of spatial domain information statistics, and provides a new method for correcting the noise of the fixed mode according to the characteristics of the noise of the fixed mode from a noise model of the fixed mode.

Description

A Correction Method for Image Fixed Pattern Noise

technical field

The invention relates to a method for correcting image fixed pattern noise, which belongs to the technical field of image processing.

Background technique

The imaging process of the camera is the process of digitizing the light signal. The light first passes through the lens to reach the photosensitive element, that is, the image sensor (CCD, CMOS, etc.), converts the light signal into a digital signal, and then passes through a special digital signal processing chip for subsequent signal processing and then transmits it to cameras and other equipment. Generally, an image sensor includes a large number of photosensitive units, and each photosensitive unit corresponds to a pixel in an image output by the image sensor.

Defects inherent in image sensors, such as inconsistent response of sensor pixel cells to imaging targets due to dark current non-uniformity and pixel defects, errors in pixel cell size due to process issues, uneven optical coating of detection devices, and readout Circuit mismatches can cause fixed pattern noise in the image.

At present, most CMOS image sensors use a column sharing structure for signal readout. In this structure, although the pixel exposure is uniform, due to process deviation, the properties of different column readout circuits are different. For different columns, The output of pixels is uneven, which leads to the generation of column fixed pattern noise, which appears as vertical stripes on the output image, which seriously affects the image quality and has a serious impact on subsequent tasks such as information extraction and target recognition. Therefore, the correction is fixed. Pattern noise is of great significance for image applications.

SUMMARY OF THE INVENTION

In order to solve the above technical problems, the present invention provides a method for correcting image fixed pattern noise, and its specific technical scheme is as follows:

A method for correcting image fixed pattern noise, comprising the following steps:

Step 1: Obtain image data: First, take the mean and standard deviation of each column of the image to obtain a row and column mean data Fmean and a row and column standard deviation data Fstd;

Step 2: Find the mean Nave: traverse all the data of the column mean Fmean, for each position in Fmean, set this position as Wmid, take Wmid as the center, and (N-1)/2 data on the left and right, there are N in total data, record the N data as Wmid window data, and average the Wmid window data Nave;

Step 3: Find the mean value NAve of the position P data: For all data greater than Nave in the Wmid window data, write down the position P in the Wmid window where these data are located, and find the mean value NAve of the position P data;

Step 4: Find the mean value of the data at position P Ivar: traverse all the data of the column standard deviation Fstd, for each position in Fstd, set this position as Wmid', with Wmid' as the center, each (N-1)/2 data, a total of N data (N is an odd number), record the N data as Wmid' window data, and find the data mean value Ivar of the location P in the Wmid' window data;

Step 5: Set the decision threshold: set the decision threshold Flim=2*NAve-Nave;

Step 6: Image correction: When the column mean value of the Wmid position Fmean(Wmid)<Flim, the image column corresponding to this position is judged as fixed pattern noise, and the image column is corrected, as shown in formula (1)

Fim(Wmid) = Ivar/Fstd(Wmid)*(F(Wmid)-Fmean(Wmid))+NAve(1)

In the formula, F(Wmid) is the image column with fixed pattern noise, Fmean(Wmid) is the mean of the image column, Fstd(Wmid) is the standard deviation of the image column, and Fim(Wmid) is the image with fixed pattern noise removed List;

Step 7: Repeated traversal: Repeat steps 2 to 6 in sequence until the image sequence traversal correction is completed.

Further, N in the steps 2 and 4 is an odd number.

Beneficial effects of the present invention:

The method for correcting image fixed pattern noise of the present invention is based on the method of spatial domain information statistics, starting from the fixed pattern noise model and according to the characteristics of the fixed pattern noise, and proposes a new fixed pattern noise correction method. The computational complexity of this method is relatively high. Other methods are not high and easy to implement, and the noise correction effect of this method is more ideal, while retaining more detailed information of the image, improving the image display effect, and providing a reference for subsequent image processing tasks such as target recognition, information extraction, etc. .

Description of drawings

Fig. 1 is the flow chart of the present invention,

Fig. 2 is the noise figure with fixed pattern of the present invention,

FIG. 3 is a corrected diagram of FIG. 2 .

Detailed ways

The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, and only illustrate the basic structure of the present invention in a schematic manner, so they only show the structures related to the present invention.

At present, there are three main categories of fixed pattern noise removal methods: methods based on spatial domain information statistics, methods based on spectral filtering, total variation and improved total variation algorithms. The methods based on spatial domain information statistics mainly include histogram matching, moving window moment matching and their improved algorithms. This type of method uses the favorable information in the original image to select a reference image column that is not affected by the fixed pattern noise in the original image, and uses the reference image column to perform histogram matching or moment matching on the fixed pattern noise image. The grayscale information solves the reference image column, resulting in a smoother image, serious loss of image details, and serious damage to the overall structure of the image. The methods based on spectral filtering mainly include joint filtering in frequency domain and spatial domain and wavelet transform filtering. However, because the frequency of noise cannot be determined, image details will be lost and the linearity of imaging blur will appear during filtering. Various algorithms based on total variation construct energy functionals with constraints, but in the process of algorithm convergence, strong constraints make the image staircase effect more serious. As shown in Figure 1, the method for correcting image fixed pattern noise of the present invention is based on the method of spatial domain information statistics, starting from the fixed pattern noise model, and according to the characteristics of the fixed pattern noise, a new fixed pattern noise correction method is proposed, Compared with other methods, the computational complexity of this method is not high, and it is easy to implement, and the noise correction effect of this method is more ideal. At the same time, it retains more detailed information of the image, improves the image display effect, and is suitable for subsequent image processing tasks such as target recognition. , information extraction, etc.

Example 1

As shown in Figure 2, first, the column mean Fmean and the column standard deviation Fstd of each column in Figure 2 are calculated. The number of image columns obtained by division in Figure 2 is 600, so Fmean and Fstd each have 600 data. Then, for each of the 600 data locations in Fmean, label this location as Wmid. If the value of N is set to 5, then there are 3 data on the left and right of the Wmid position, and there are 7 data in Wmid itself, and the average Nave of these 7 data is obtained. Next, find out all the data larger than Nave in the above 7 data, make the position of the data larger than Nave to be P, and obtain the mean value NAve of the data at the position P of the data larger than Nave. For each position of the 600 data in Fstd, this position is marked as Wmid', and there are 3 data on the left and right of the Wmid' position, for a total of 7 data. Similarly, there is the corresponding position P in step 3 in these 7 data, find the mean value Ivar of position P in Wmid'. After calculating NAve and Ivar, set the decision threshold so that Flim=2*NAve-Nave. When the column mean of the Wmid position is Fmean(Wmid)<Flim, the image column corresponding to this position is judged as fixed pattern noise, and the image column at this position is corrected, as shown in formula (1),

Fim(Wmid) = Ivar/Fstd(Wmid)*(F(Wmid)-Fmean(Wmid))+NAve(1)

In the formula, F(Wmid) is the image column with fixed pattern noise, Fmean(Wmid) is the mean of the image column, Fstd(Wmid) is the standard deviation of the image column, and Fim(Wmid) is the image with fixed pattern noise removed List. In the same way, all 600 image columns in the image are traversed one by one, and NAve and Ivar are obtained for threshold judgment. All image columns with fixed pattern noise in Figure 2 will be corrected and removed at the end, and then the corrected result image will be obtained. ,As shown in Figure 3.

The method for correcting image fixed pattern noise of the present invention is based on the method of spatial domain information statistics, starting from the fixed pattern noise model and according to the characteristics of the fixed pattern noise, and proposes a new fixed pattern noise correction method. The computational complexity of this method is relatively high. Other methods are not high and easy to implement, and the noise correction effect of this method is more ideal, while retaining more detailed information of the image, improving the image display effect, and providing a reference for subsequent image processing tasks such as target recognition, information extraction, etc. .

Taking the above ideal embodiments according to the present invention as inspiration, and through the above description, relevant personnel can make various changes and modifications without departing from the technical idea of the present invention. The technical scope of the present invention is not limited to the contents in the specification, and the technical scope must be determined according to the scope of the claims.

Claims (2)

1. a correction method of image fixed pattern noise is characterized in that: comprise the following steps: Step 1: Obtain image data: First, take the mean and standard deviation of each column of the image to obtain a row and column mean data Fmean and a row and column standard deviation data Fstd; Step 2: Find the mean Nave: traverse all the data of the column mean Fmean, for each position in Fmean, set this position as Wmid, take Wmid as the center, and (N-1)/2 data on the left and right, there are N in total data, record the N data as Wmid window data, and average the Wmid window data Nave; Step 3: Find the mean value NAve of the position P data: For all data greater than Nave in the Wmid window data, write down the position P in the Wmid window where these data are located, and find the mean value NAve of the position P data; Step 4: Find the mean value of the data at position P Ivar: traverse all the data of the column standard deviation Fstd, for each position in Fstd, set this position as Wmid', with Wmid' as the center, each (N-1)/2 data, a total of N data (N is an odd number), record the N data as Wmid' window data, and find the data mean value Ivar of the location P in the Wmid' window data; Step 5: Set the decision threshold: set the decision threshold Flim=2*NAve-Nave; Step 6: Image correction: When the column mean value of the Wmid position Fmean(Wmid)<Flim, the image column corresponding to this position is judged as fixed pattern noise, and the image column is corrected, as shown in formula (1) Fim(Wmid) = Ivar/Fstd(Wmid)*(F(Wmid)-Fmean(Wmid))+NAve(1) In the formula, F(Wmid) is the image column with fixed pattern noise, Fmean(Wmid) is the mean of the image column, Fstd(Wmid) is the standard deviation of the image column, and Fim(Wmid) is the image with fixed pattern noise removed List; Step 7: Repeated traversal: Repeat steps 2 to 6 in sequence until the image sequence traversal correction is completed. 2 . The method for correcting fixed pattern noise of an image according to claim 1 , wherein N in the steps 2 and 4 is an odd number. 3 .

Images