Composite correction method for spatial heterogeneity between pixels

Abstract

The invention discloses a composite correction method for spatial heterogeneity between pixels, and belongs to the technical field of semiconductors. The method is applied to a pixel array formed by a composite dielectric grating photosensitive detector, and comprises the following steps of: fitting a light response curve model of each pixel in sections according to the characteristics of the composite dielectric grating photosensitive detector; a multi-point uniform field correction method is adopted to correct data collected by a pixel array composed of composite dielectric grating photosensitive detectors, and a single-point correction method is adopted to correct abnormal conditions. According to the scheme, not only can the DRNU be well corrected, but also the PRNU can be well corrected, meanwhile, the problem of nonlinearity when low-light imaging and close to a full well exist in a light response curve of the composite dielectric gate light-sensitive detector can be solved, and further, the number of fitting segments can be reduced or increased according to the cache capacity of the detector itself, and whether abnormal conditions are abandoned or not can be judged.

Description

Technical Field

[0001] The invention relates to a composite correction method for spatial non-uniformity between pixels, belonging to the technical field of semiconductors. Background Art

[0002] In the field of digital image processing and computer vision, when capturing images, sensor devices (such as cameras) often have inconsistent responses between different pixels due to factors such as hardware manufacturing errors, sensor noise, optical distortion, etc. This inconsistency between pixels can cause stripe noise and fixed pattern noise (FPN) in the image, causing image quality degradation, especially in fine image processing tasks such as high-precision imaging, medical image processing, remote sensing imaging, etc.

[0003] In terms of the types of specific characteristic parameters that differ between pixels of photodetector imaging chips, non-uniformity usually manifests itself in the following two ways: DSNU (Dark Signal Nonuniformity), that is, the different dark currents of each pixel in the dark state lead to inconsistent grayscale values ​​obtained by imaging, which corresponds to the intercept term in the light response formula; PRNU (Photo Response Nonuniformity), that is, the inconsistency of the response of each pixel to light under the same lighting conditions, which corresponds to the slope term in the light response formula.

[0004] Regarding the fixed pattern noise FPN mentioned above, it presents a fixed spatial pattern in each pixel of the image sensor, such as brightness or color deviation at a fixed position. The commonly used methods for correcting the FPN caused by this spatial non-uniformity are black field correction and two-point mean field correction. Among them, although the black field correction is convenient to operate, it only needs to collect a dark field picture once for subtraction, but it can only solve the problem of inconsistency of initial grayscale values ​​between pixels; the two-point mean field correction determines the gain and offset of each pixel through two reference images, and then corrects the gain and offset of each pixel. This method can correct DSNU and PRNU at the same time, but the light response curve of the photosensitive detector is not a very strict linear curve, and cannot handle nonlinear situations when it is close to the full well situation.

[0005] A Chinese invention patent with publication number CN102938409A proposes a composite dielectric grating photosensitive detector, which is characterized in that a single semiconductor device can realize complete reset, photosensitive and readout functions, forming a complete pixel, which can greatly improve the pixel fill factor. This composite dielectric grating photosensitive detector has more full well charges due to the presence of two photoelectron storage nodes in its structure, and the existing non-uniformity correction method cannot handle the nonlinear situation close to the full well situation, so the problem caused by the inconsistency between pixels on this detector will be more obvious. Summary of the invention

[0006] In order to solve the existing problems, the present invention provides a composite correction method for the spatial non-uniformity between pixels of a composite dielectric grating photosensitive detector, comprising:

[0007] Step S1, fitting a light response curve model of each pixel by a piecewise fitting method;

[0008] Step S2, according to the light response curve model data, a multi-point mean field correction method is used to correct the data collected by the pixel array composed of the composite dielectric grating photosensitive detector;

[0009] Step S3, determining whether the fixed pattern noise corresponding to the data corrected by the multi-point average field correction method exceeds a preset DN threshold;

[0010] Step S4: If the fixed pattern noise corresponding to the corrected data still exceeds the preset DN threshold, a single-point mean field correction method is further used for correction.

[0011] Optionally, the step S1 is to fit the light response curve model of each pixel using a three-segment fitting method, including:

[0012] Step S1.1, set the integration time, use uniform light to illuminate the surface of each detector, control the light intensity to gradually increase from 0 to a value that just makes the detector full well, and collect M pictures at each light intensity point;

[0013] Step S1.2, generating average data of each light intensity point based on the image collected in step S1.1;

[0014] Step S1.3, taking the data corresponding to the light intensity of 0 as the dark field data, and subtracting the dark field data from the data of other light intensity points to generate pure light response data;

[0015] Step S1.4, performing a three-segment fitting of the light response curve based on the pure light response data, determining the slope and intercept of each segment, and then obtaining a light response curve after a fitting according to the slope and intercept of each segment curve.

[0016] Optionally, when performing three-segment fitting of the light response curve in step S1.4, the basic linear light response curve formula is:

[0017]

[0018] Among them, m1, m2, and m3 are the slopes of the three segments, b1, b2, and b3 are the intercepts of the three segments, photon1 and photon2 are the horizontal coordinates of the intersections of the three segments; Photon represents the light intensity value, and DN represents the pixel value.

[0019] Optionally, when performing the three-stage fitting of the light response curve in step S1.4, the fitting is performed for each pixel according to the following conditions:

[0020] make Minimum

[0021] make Minimum

[0022] make Minimum

[0023] Among them, DN i Indicates the pixel value corresponding to the i-th light intensity, Photon i Represents the i-th light intensity value.

[0024] Optionally, the step S2 uses a multi-point mean field correction method to correct the data collected by the pixel array composed of the composite dielectric grating photosensitive detector according to the light response curve model data, including:

[0025] Step S2.1, obtaining the image data to be corrected collected by the pixel array composed of the composite dielectric grating photosensitive detector, and subtracting the dark field image data actually shot with the same integration time to obtain the pure light response data;

[0026] Step S2.2, determine the position of each pixel on the light response curve after the first fitting, and perform correction according to its position. The corrected DN value is recorded as DN correct , calculated as follows:

[0027]

[0028] DN correct =K*Photon

[0029] Among them, DN data is the DN value of any pixel, the slope corresponding to the position of any pixel on the light response curve after a fitting is recorded as m, and the intercept is recorded as b; m is any value among m1, m2, and m3, and b is any value among b1, b2, and b3; mean data_DN is the mean DN value of the image to be corrected, mean Photon The DN value of each pixel in the image to be corrected is converted into the average value of the Photon value through the light response curve.

[0030] Optionally, step S4 includes:

[0031] If the fixed pattern noise corresponding to the corrected data still exceeds the preset DN threshold, the DN value point closest to the DN value is found on the light response curve after the first fitting, which is DN.model , the corrected DN value DN correct The specific calculation is as follows:

[0032] DN correct =mean model -(DN model -DN data )

[0033] Among them, mean model Indicates the mean DN value of the image at the selected DN value point.

[0034] Optionally, the preset DN threshold is an empirical value.

[0035] The beneficial effects of the present invention are:

[0036] The present invention is aimed at a pixel array composed of composite dielectric grating photosensitive detectors, and designs a composite correction method to solve the problem of spatial non-uniformity. The light response curve model of each pixel is segmented and fitted according to its characteristics. According to the light response curve model data, a multi-point mean field correction method is used to correct the data collected by the pixel array composed of composite dielectric grating photosensitive detectors, and a single-point correction method is used for abnormal situations. This scheme can not only correct DRNU well, but also PRNU well, and can solve the problem of nonlinearity in low-light imaging and near full well in the light response curve of the composite dielectric grating photosensitive detector. Furthermore, according to the cache capacity of the detector itself, the number of fitting segments can be reduced or increased, and whether to discard abnormal situations can be reduced or increased. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 A flow chart of a composite correction method for spatial non-uniformity between pixels provided in the present application.

[0039] Figure 2 Schematic diagram of the structure of the composite dielectric grating photosensitive detector in the embodiment.

[0040] Figure 3 It is the actual light response curve of the composite dielectric grating photosensitive detector.

[0041] Figure 4 It is the relationship between the FPN and the mean DN value of the image after using the multi-point average field method to process the results of shooting under different light intensities (uniform light).

[0042] Figure 5 It is the relationship between FPN and the mean DN value of the image after post-processing the results taken under different light intensities (uniform light) using the composite method.

[0043] Figure 6 This is a comparison of images before and after correcting spatial non-uniformity using the composite correction method and the black field correction method. The left image is after correction using the composite correction method, and the right image is after correction using the black field correction method.

[0044] Figure 7 This is an image comparison diagram when using the composite correction method and the two-point average field correction method to correct certain abnormal DN value ranges. The left image is after correction using the composite correction method, and the right image is after correction using the two-point average field correction method. DETAILED DESCRIPTION

[0045] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0046] Embodiment 1:

[0047] This embodiment provides a composite correction method for spatial non-uniformity between pixels, such as Figure 1 As shown, the method is applied to a pixel array composed of a composite dielectric grating photosensitive detector, comprising:

[0048] Step S1, fitting a light response curve model of each pixel by a piecewise fitting method;

[0049] Step S2, according to the light response curve model data, a multi-point mean field correction method is used to correct the data collected by the pixel array composed of the composite dielectric grating photosensitive detector;

[0050] Step S3, determining whether the fixed pattern noise corresponding to the data corrected by the multi-point average field correction method exceeds a preset DN threshold;

[0051] Step S4: If the fixed pattern noise corresponding to the corrected data still exceeds the preset DN threshold, a single-point mean field correction method is further used for correction.

[0052] The method is based on a stable uniform light source generator, which is used for lighting when collecting and generating model data.

[0053] The composite dielectric gate photosensitive detector constituting the pixel array is a composite dielectric gate MOSFET photosensitive detector disclosed in the PCT patent application with publication number WO2010 / 094233A1, such as Figure 2As shown, the composite dielectric grating photosensitive detector includes a MOS-C part and a MOSFET part, and the two parts are formed on the same P-type semiconductor substrate. Among them, the MOS-C part includes a first dielectric layer, a charge coupling layer, a second dielectric layer and a first control gate stacked in sequence on the P-type semiconductor substrate; the MOSFET part includes a first dielectric layer, a charge coupling layer, a second top dielectric layer and a second control gate stacked in sequence on the P-type semiconductor substrate, an N-type source region and an N-type drain region are provided on one side of the P-type semiconductor substrate and close to the bottom dielectric layer, and a threshold adjustment injection region is provided in the P-type semiconductor substrate and below the bottom dielectric layer.

[0054] The composite correction method for the spatial non-uniformity between pixels of the composite dielectric grating photosensitive detector specifically includes:

[0055] The first step is to use a uniform light source to generate uniform light to illuminate the detector surface, set a suitable integration time, and change the light intensity of the uniform light source, increasing it step by step from no light to the light intensity that just fills the detector well, and collect 10 images at each light intensity point;

[0056] In the case of no illumination, that is, the light intensity is 0, assuming that there are n light intensities from 0 to the light intensity that just fills the detector well, then 10N images are collected. The integration time can be set based on experience, such as 30ms or 40ms.

[0057] The second step is to generate average data of each light intensity point based on the image data collected in the first step;

[0058] For each pixel, the corresponding pixel values ​​of 10 images under the same light intensity are averaged, and the average value is used as the average data of the light intensity point.

[0059] The third step is to treat the data shot without light as a dark field, and subtract the dark field from the data of each other light intensity point to generate pure light response data;

[0060] The fourth step is to perform a three-segment fitting of the light response curve based on the pure light response data and determine the slope and intercept of each segment;

[0061] like Figure 3 As shown, the light response curve of the composite dielectric grating photosensitive detector presents a three-segment characteristic, and the three-segment fitting light response curve is in the form of:

[0062]

[0063] Among them, m1, m2, m3 are the slopes of the three segments, b1, b2, b3 are the intercepts of the three segments, photon1 and photon2 are the horizontal coordinates of the intersection points of the three segments;

[0064] When performing a three-stage fitting of the light response curve, the fitting is performed according to the following conditions, taking a single pixel as an example:

[0065] make Minimum

[0066] make Minimum

[0067] make Minimum

[0068] Among them, DN i Indicates the pixel value corresponding to the i-th light intensity, Photon i represents the i-th light intensity value;

[0069] The slope (m1, m2, m3) and intercept (b1, b2, b3) of each segment of the first-order curve are stored, and the light response curve after first-order fitting is obtained according to the slope and intercept of each segment of the curve;

[0070] The fifth step is to restore the light response curve model of each pixel based on the slope (m1, m2, m3) and intercept (b1, b2, b3) of each segment of the linear curve determined in the fourth step, using the multi-point mean field correction method:

[0071] Step 6: Collect the image to be corrected. First, subtract the real image without lighting taken with the same integration time. Take any pixel in the image as an example, and its DN value is DN data , find the corresponding light response curve after a single fitting, with a slope of m and an intercept of b. The mean DN value of the image to be corrected on the light response curve after a single fitting is mean data_DN The DN value of each pixel in the image to be corrected is converted into the mean value of Photon through the light response curve. photon , then the corrected DN value is DN correct The specific calculation is as follows:

[0072]

[0073] DN correct =K*Photon

[0074] Wherein, K is the conversion coefficient from Photon to DN, and Photon is the light intensity value corresponding to this pixel according to the light response curve.

[0075] Step 7: Process the results of shooting under different light intensities according to the method in step 6 to find out the abnormal points. The abnormal points are the pixels whose fixed pattern noise is still greater than the number of electrons represented by 1 DN value after the above correction. The number of electrons represented by 1 DN value is determined according to the specific device, such as 40 electrons. Figure 4 shown.

[0076] Step 8: Use the single-point mean field method at these outliers:

[0077] According to the pure light response data generated in the third step, for any pixel in the collected image to be corrected, its DN value is DN data , you can find the DN value point closest to it in the model data, which is DN model , the corrected DN value DN correct The specific calculation is as follows

[0078] DN correct =mean model -(DN model -DN data )

[0079] Among them, mean model Indicates the mean DN value of the image at the selected DN value point.

[0080] The results of the composite correction method are as follows Figure 5 shown.

[0081] Finally, the results of the composite correction method and the black field correction method proposed in the present invention are compared. Figure 6 As shown in the figure, it can be seen that the noise of the structure fixed pattern processed by the black field correction method is too large. This is because it does not correct the slope inconsistency of the light response and in some abnormal DN value ranges; the results of using the multi-point average field method and the single-point average field method are compared, as shown in Figure 7 As shown, it can be seen that the results of the single-point average field method in the abnormal DN value range have a lot of abrupt noise, which is also fixed pattern noise that has not been properly corrected.

[0082] Some steps in the embodiments of the present invention may be implemented using software, and the corresponding software program may be stored in a readable storage medium, such as a CD or a hard disk.

[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A composite correction method for spatial non-uniformity between pixels, characterized in that: The method is applied to a pixel array composed of a composite dielectric grating photosensitive detector, and the method comprises: Step S1, fitting a light response curve model of each pixel by a piecewise fitting method; Step S2, according to the light response curve model data, a multi-point mean field correction method is used to correct the data collected by the pixel array composed of the composite dielectric grating photosensitive detector; Step S3, determining whether the fixed pattern noise corresponding to the data corrected by the multi-point average field correction method exceeds a preset DN threshold; Step S4: If the fixed pattern noise corresponding to the corrected data still exceeds the preset DN threshold, a single-point mean field correction method is further used for correction.

2. The method according to claim 1, characterized in that The step S1 is to fit the light response curve model of each pixel using a three-segment fitting method, including: Step S1.1, set the integration time, use uniform light to illuminate the surface of each detector, control the light intensity to gradually increase from 0 to a value that just makes the detector full well, and collect M pictures at each light intensity point; Step S1.2, generating average data of each light intensity point based on the image collected in step S1.1; Step S1.3, taking the data corresponding to the light intensity of 0 as the dark field data, and subtracting the dark field data from the data of other light intensity points to generate pure light response data; Step S1.4, performing a three-segment fitting of the light response curve based on the pure light response data, determining the slope and intercept of each segment, and then obtaining a light response curve after a fitting according to the slope and intercept of each segment curve.

3. The method according to claim 2, characterized in that When the step S1.4 performs three-stage fitting of the light response curve, the basic linear light response curve formula is: Among them, m1, m2, and m3 are the slopes of the three segments, b1, b2, and b3 are the intercepts of the three segments, photon1 and photon2 are the horizontal coordinates of the intersections of the three segments; Photon represents the light intensity value, and DN represents the pixel value.

4. The method according to claim 3, characterized in that When the three-stage light response curve is fitted in step S1.4, the fitting is performed for each pixel according to the following conditions: Among them, DN i Indicates the pixel value corresponding to the i-th light intensity, Photon i Represents the i-th light intensity value.

5. The method according to claim 4, characterized in that The step S2 uses a multi-point mean field correction method to correct the data collected by the pixel array composed of the composite dielectric grating photosensitive detector according to the light response curve model data, including: Step S2.1, obtaining the image data to be corrected collected by the pixel array composed of the composite dielectric grating photosensitive detector, and subtracting the dark field image data actually shot with the same integration time to obtain the pure light response data; Step S2.2, determine the position of each pixel on the light response curve after the first fitting, and perform correction according to its position. The corrected DN value is recorded as DN correct , calculated as follows: DNcorrecctt=K*Photon Among them, DN data is the DN value of any pixel, the slope corresponding to the position of any pixel on the light response curve after a fitting is recorded as m, and the intercept is recorded as b; m is any value among m1, m2, and m3, and b is any value among b1, b2, and b3; mean data _ DN is the mean DN value of the image to be corrected, mean Photon The DN value of each pixel in the image to be corrected is converted into the average value of the Photon value through the light response curve.

6. The method according to claim 5, characterized in that The step S4 comprises: If the fixed pattern noise corresponding to the corrected data still exceeds the preset DN threshold, the DN value point closest to the DN value is found on the light response curve after the first fitting, which is DN. model , the corrected DN value DN correct The specific calculation is as follows: DNcorrect=meanmodel-(DNmodel-DNdata) Among them, mean model Indicates the mean DN value of the image at the selected DN value point.

7. The method according to claim 6, characterized in that The preset DN threshold is an empirical value.

Images