CN110012287B - The dirty self checking method of digital camera image sensor based on retina perception - Google Patents
The dirty self checking method of digital camera image sensor based on retina perception Download PDFInfo
- Publication number
- CN110012287B CN110012287B CN201910118410.9A CN201910118410A CN110012287B CN 110012287 B CN110012287 B CN 110012287B CN 201910118410 A CN201910118410 A CN 201910118410A CN 110012287 B CN110012287 B CN 110012287B
- Authority
- CN
- China
- Prior art keywords
- digital camera
- image sensor
- data
- dirty
- camera image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N17/00—Diagnosis, testing or measuring for television systems or their details
- H04N17/002—Diagnosis, testing or measuring for television systems or their details for television cameras
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of dirty self checking methods of digital camera image sensor based on retina perception, contain following steps: 1, placing a pure color object before digital camera camera lens, take pictures, obtain the image data of a frame RGB channel；2, it reads in image data and carries out transcendental method gray processing；3, a convolution kernel is set, it is down-sampled to carry out the mean value based on convolution；4, convolutional calculation and calculating from quotient images based on retina perception, obtain the data of relocatable；5, data processing is carried out；6, crosscorrelation template matching is normalized using normalization cross correlation algorithm, obtains correlation temperature figure；7, thresholding processing is carried out to correlation temperature figure；8, connected domain is analyzed, the dirty position of digital camera image sensor is obtained.The present invention can fast and effeciently extract itself dirty feature of digital camera image sensor and detect, and testing cost is low, and processing speed is fast, it is easy to accomplish.
Description
(1), technical field:
The present invention relates to a kind of detection methods of image sensor, and in particular to a kind of digital phase based on retina perception
The dirty self checking method of machine image sensor.
(2), background technique:
Image enchancing method is in automatic Pilot, outdoor robot, aerospace, biomedicine, public safety, AR, VR, work
Intelligence manufacture in industry etc. is essential item of image processing step.Before computer vision algorithms make is formally handled,
Algorithm for image enhancement will be carried out to original image to improve the part details in image, such as progress tone remaps, edge enhancing
It is all based on Deng, this method and stretches the lesser characteristic of script codomain range into a biggish dynamic range.
Most of real time algorithms are all based on artificial modeling, therefore algorithm needs to perceive global gray scale, such as ash
Spend thresholding, HOG, clustering, FSAT feature etc..Due to the needs of image enhancement, when camera sensor is in poor work
When having adhered to the tiny sundries for falling into body after working long hours in environment, on sensor, the blackening of imaging is handled with enhancing
It further highlights, directly affects the reliability even availability of algorithm.
The dirty of camera sensor is all high-precisions, the hidden danger of high-tech view-based access control model intelligently equipped, for bright
Aobvious is dirty, can by human eye or traditional detection algorithm identification sensor whether by dirty influence, and it is some light and
It is small dirty more to influence vision system, it is generally the case that it is significantly dirty that very deep blackening can be obtained by processing, easily
In identification, and thin dirty, can be mingled in the gray feature of acquisition target, cause detection, the failure of tracking scheduling algorithm or
It is unstable.
There are two types of the dirty detection technique of existing camera sensor is general, one is detected sensings of being taken pictures by peripheral hardware
Device surface belongs to theoretical method, practical application meaning is lower because general detection environment is unable to satisfy dustless in this way；
Another kind is the self-checking without carrying out disassembling section, and the discrimination of the technology is lower, is primarily due to lens imaging principle
It will lead to exposure and unevenly halation phenomenon occur, show the dark effect of intermediate bright surrounding, can not operate by thresholding have
Effect is partitioned into unconspicuous blackening.
And the vision of the mankind has the characteristics that the adaptive of complexity, being mainly manifested in has centainly slowly varying light environment
Fitting or filtering feature, it is most of dirty to pass through some increasings therefore when using the detection scheme of manual method
Strong algorithms are observed, but are difficult to carry out effective automatic detection directly against the result of enhancing, essentially consist in different types of light
Source can not all overcome the image in incident camera lens to have halation, and the result of enhancing still remains with the effect (as shown in Figure 1) of halation.
(3), summary of the invention:
The technical problem to be solved by the present invention is providing a kind of digital camera image sensor based on retina perception
Dirty self checking method, the self checking method can fast and effeciently extract itself dirty feature of digital camera image sensor and examine
It surveys, testing cost is low, and processing speed is fast, it is easy to accomplish.
Technical solution of the present invention:
A kind of dirty self checking method of the digital camera image sensor based on retina perception, contains following steps:
Step 1: placing a pure color object before digital camera camera lens, which is completely covered digital camera mirror
The head visual field, is taken pictures with digital camera, obtains the image data of a frame RGB channel；
Step 2: carrying out transcendental method gray processing to the image data；
Step 3: one convolution kernel of setting, it is down-sampled to carry out the mean value based on convolution；
Step 4: carrying out convolutional calculation and calculating from quotient images (SQI) based on retina perception, relocatable is obtained
Data；
Step 5: carrying out data processing, no symbol shape data is obtained；
Step 6: crosscorrelation template matching is normalized using normalization crosscorrelation (NCC) algorithmic formula, and lead to
Sliding window mode is crossed, correlation temperature figure (HeatMap) is obtained；
Step 7: carrying out thresholding processing to correlation temperature figure；
Step 8: analysis connected domain, obtains the dirty position of digital camera image sensor.
In the transcendental method of step 2, gray level image, first proved recipe are obtained using the mode being maximized to RGB channel component
Method is nonlinear processing function, and the function expression is as follows:
H (x, y)=MAX (R (x, y), G (x, y), B (x, y))
Wherein, H (x, y) is the gray-scale pixel values at position (x, y), and MAX (R (x, y), G (x, y), B (x, y)) is to ask three
The function of maximum value in a input value, R (x, y), G (x, y), B (x, y) respectively represent RGB in the pixel at position (x, y)
Channel components value.
The problem of gray processing brightness caused by the transcendental method of Nonlinear Processing overcomes because of white balance reduces.
Convolution kernel format in step 3 are as follows:One pixel is one volume 3 × 3 corresponding
Product assess calculation as a result, the down-sampled sliding window of mean value with step-length (3,3) for stepping.
Using the down-sampled technology based on convolution method, subsequent place is accelerated while overcoming noise and moire fringes effect
Manage speed.
Sampling function S is designed, is calculated with convolution kernel, obtains pixel value, convolution kernel is a kind of square window, inside
Portion uses average calculation method.
Step 4 uses the relocatable that sensing results are calculated from quotient images (SQI) using filtering convolutional calculation
Data.
Contain a kind of retina sensor model from quotient images (SQI) calculating, passes through the Flanker task and vision of neuron
Mach band effect, to imitate the Mach band effect in human vision.
It is a kind of method from quotient images, essentially consists in original image and obtain new image after treatment, with the picture of new image
Element value carries out division arithmetic as the pixel value of denominator and original image, and when operation is that operation is executed with individual element, calculates every time
New result as output image corresponding position pixel value.
Use extraneous pure color object that can make to expose enough by area source or any brightness as the illumination source of detection
Object in normal range (NR) is realized, by Lambert illumination model, can analyze its BRDF function.
Spherical coordinate system is set in three-dimensional world, the energy of external light source incidence is I (λ, θi,φi), reflected energy be R (λ,
θr,φr), then object reflecting attribute s (λ, θi,φi,θr,φr) expression formula can be written as:
Wherein, θ, φ are respectively incidence/reflection light angle, and λ is then spectral information, define the color point of light
Amount.
In view of diffusing reflection object, variable abbreviation that key property enables BRDF function to be analyzed are as follows:
Considered with the modeling of human eye light, it is as follows to design a kind of simplified model:
R (λ)=s (λ) I (λ)
R (λ) is data of the human eye by light perception, is a kind of two-dimensional matrix that numerical value indicates actually, and object reflecting attribute
Numerical value indicate and illumination attribute numerical value expression be that one kind multiplies sexual intercourse.
Again because the photobehavior of camera is a kind of simple exposure attribute, above-mentioned expression formula is directly applied, it can be with
Obtain the expression formula of SQI:
Here, I (x, y) is original exposure data, a kind of expression of L (x, y) light brightness distribution refers in particular to one kind and diffuses
Numeralization according to model indicates.
External object is a kind of diffusing reflection object, meets Lambert illumination model.Taken the photograph object and camera lens halation are all inherited
The property of planar light source.
Therefore, all information of exposure can be seen as diffusing reflection object, that is to say, that from quotient images (SQI) be very
It is succinct but scientific and effective.
Further, a kind of method is designed, so that I (x, y) relatively accurately estimates a kind of light when there is dirty information
According to modeling data L (x, y), it is thus evident that the data for using original image I (x, y) select a kind of filtering method that can just obtain illumination modeling
Data L (x, y).
Filtering method needs to select in conjunction with Mach band effect, is realized using Gaussian filter.
Further, it needs to be detected according to sensing results R (x, y).
Conversion calculating of the step 5 using floating-point to unsigned int data, concrete operation step are as follows: to from quotient images
(SQI) calculated result carries out -0.5 grey scale pixel value offset, then will not be set to 0.5 gray scale in the data of codomain [0,1], finally
The numerical value zoom operations that 255 times are carried out to whole image pixel value, obtain 8bit without symbol shaping grey scale pixel value.
Before treatment in order to accelerate to calculate, need to handle the real-coded GA of sensing results for shape data, this hair
The bright middle shape data for selecting 8bit precision, therefore have following pretreatment:
It is greater than 1 and the data less than 1 in SQI first of all for retaining, the increment of -0.5 gray scale is carried out to whole pixel brightness,
Weaken the amplitude of 0.5 gray scale, filter codomain not in the value of [0,1], be set as 0.5 gray scale:
R (x, y)=R (x, y) -0.5
If (R (x, y) < 0) R (x, y)=0.5
If (R (x, y) > 1) R (x, y)=0.5
Finally zoom in and out processing:
R ' (x, y)=R (x, y) * 255.0
R ' (x, y)=Round (R ' (x, y))
Wherein Round function plays the role of rounding up and switchs to shaping.
Detection method used in step 6 is known as NCC (Normalized Cross Correlation), that is, normalizes
Whether crosscorrelation can will have relationship quantization, often use in big data analysis between data sequence and sequence.About original
Definition, a two-dimensional sequence I1Certain region and another template ItThe degree of correlation can be indicated with following form:
Mean information is respectively as follows:
Complete comparison type is detected, that is, sliding window detection, is had:
Wherein:
In formula, the matrix of (x, y) ∈ [M*N], [m*n] can indicate window size.
Eventually by sliding window mode, a kind of temperature figure based on normalization crosscorrelation property has been obtained.
Matching template comes from pre-training library.
For multiple dirty, temperature figure can become more chaotic, but have no effect on the normal output of self-test, because when not having
When dirty, temperature figure is more clean and tidy and not in threshold range.
Preferably, pure color object is area source in step 1.
Preferably, area source is planar light source.
Beneficial effects of the present invention:
1, present invention uses retina perception theories, slowly varying to illumination insensitive when being perceived using human eye retina
Characteristic perceived using retina is established from quotient images method by the Flanker task and vision Mach band effect of neuron
Model imitates the Mach band effect in human vision, design it is simple and effective from quotient images SQI frame, set relative to traditional
Standby detection and artificial detection, the present invention can fast and effeciently extract the dirty feature of digital camera image sensor and detect, nothing
Fit procedure is needed, the speed and accuracy of human eye perception are had more than.
2, what gray processing brightness reduced caused by the transcendental method of Nonlinear Processing overcomes because of white balance in the present invention asks
Topic accelerates subsequent processing speed using the down-sampled technology based on convolution method while overcoming noise and moire fringes effect
Degree.
3, algorithm of the invention can in the form of software real time execution in embedded device, x86 compatible, server etc.
It is lower to the performance requirement of processor on programmable carrier, it is easy to accomplish.
4, the present invention need to only add pure color object (test pure color plate or planar light source can be used), not need other additional
Facility, testing cost are low.
(4), Detailed description of the invention:
Fig. 1 is image enhancement effects schematic diagram in the prior art；
Fig. 2 is the method schematic diagram with digital camera camera plane light source；
Fig. 3 is the down-sampled method schematic diagram of mean value based on convolution；
Fig. 4 is that retina perceives schematic diagram；
Fig. 5 is dirty sensing results schematic diagram；
Fig. 6 is the dirty position view after final process.
(5), specific embodiment:
The dirty self checking method of digital camera image sensor based on retina perception contains following steps:
Step 1: placing a planar light source before digital camera camera lens, which is completely covered digital camera mirror
Head the visual field, taken pictures with digital camera, obtain a frame RGB channel image data (as shown in Figure 2: 1 is planar light source；2 are
The lens set of digital camera；3 be the image sensor of digital camera)；
Step 2: carrying out transcendental method gray processing to the image data；
Step 3: one convolution kernel of setting, it is down-sampled (as shown in Figure 3) to carry out the mean value based on convolution；
Step 4: carrying out convolutional calculation and calculating from quotient images (SQI) based on retina perception, relocatable is obtained
Data；
Step 5: carrying out data processing, no symbol shape data is obtained；
Step 6: crosscorrelation template matching is normalized using normalization crosscorrelation (NCC) algorithmic formula, and lead to
Sliding window mode is crossed, correlation temperature figure (HeatMap) is obtained；
Step 7: carrying out thresholding processing to correlation temperature figure；
Step 8: analysis connected domain, obtains the dirty position of digital camera image sensor.
In the transcendental method of step 2, gray level image, first proved recipe are obtained using the mode being maximized to RGB channel component
Method is nonlinear processing function, and the function expression is as follows:
H (x, y)=MAX (R (x, y), G (x, y), B (x, y))
Wherein, H (x, y) is the gray-scale pixel values at position (x, y), and MAX (R (x, y), G (x, y), B (x, y)) is to ask three
The function of maximum value in a input value, R (x, y), G (x, y), B (x, y) respectively represent RGB in the pixel at position (x, y)
Channel components value.
The problem of gray processing brightness caused by the transcendental method of Nonlinear Processing overcomes because of white balance reduces.
Convolution kernel format in step 3 are as follows:One pixel is one volume 3 × 3 corresponding
Product assess calculation as a result, the down-sampled sliding window of mean value with step-length (3,3) for stepping.
Using the down-sampled technology based on convolution method, subsequent place is accelerated while overcoming noise and moire fringes effect
Manage speed.
Sampling function S is designed, is calculated with convolution kernel, obtains pixel value, convolution kernel is a kind of square window, inside
Portion uses average calculation method.
Step 4 uses the relocatable that sensing results are calculated from quotient images (SQI) using filtering convolutional calculation
Data.
Contain a kind of retina sensor model from quotient images (SQI) calculating, passes through the Flanker task and vision of neuron
Mach band effect, to imitate the Mach band effect in human vision (as shown in Figure 4).
It is a kind of method from quotient images, essentially consists in original image and obtain new image after treatment, with the picture of new image
Element value carries out division arithmetic as the pixel value of denominator and original image, and when operation is that operation is executed with individual element, calculates every time
New result as output image corresponding position pixel value.
Use extraneous pure color object that can make to expose enough by area source or any brightness as the illumination source of detection
Object in normal range (NR) is realized, by Lambert illumination model, can analyze its BRDF function.
Spherical coordinate system is set in three-dimensional world, the energy of external light source incidence is I (λ, θi,φi), reflected energy be R (λ,
θr,φr), then object reflecting attribute s (λ, θi,φi,θr,φr) expression formula can be written as:
Wherein, θ, φ are respectively incidence/reflection light angle, and λ is then spectral information, define the color point of light
Amount.
In view of diffusing reflection object, variable abbreviation that key property enables BRDF function to be analyzed are as follows:
Considered with the modeling of human eye light, it is as follows to design a kind of simplified model:
R (λ)=s (λ) I (λ)
R (λ) is data of the human eye by light perception, is a kind of two-dimensional matrix that numerical value indicates actually, and object reflecting attribute
Numerical value indicate and illumination attribute numerical value expression be that one kind multiplies sexual intercourse.
Again because the photobehavior of camera is a kind of simple exposure attribute, above-mentioned expression formula is directly applied, it can be with
Obtain the expression formula of SQI:
Here, I (x, y) is original exposure data, a kind of expression of L (x, y) light brightness distribution refers in particular to one kind and diffuses
Numeralization according to model indicates.
External object is a kind of diffusing reflection object, meets Lambert illumination model.Taken the photograph object and camera lens halation are all inherited
The property of planar light source.
Therefore, all information of exposure can be seen as diffusing reflection object, that is to say, that from quotient images (SQI) be very
It is succinct but scientific and effective.
Further, a kind of method is designed, so that I (x, y) relatively accurately estimates a kind of light when there is dirty information
According to modeling data L (x, y), it is thus evident that the data for using original image I (x, y) select a kind of filtering method that can just obtain illumination modeling
Data L (x, y).
Filtering method needs to select in conjunction with Mach band effect, is realized using Gaussian filter, obtains dirty perception
As a result (as shown in Figure 5).
Further, it needs to be detected according to sensing results R (x, y).
Conversion calculating of the step 5 using floating-point to unsigned int data, concrete operation step are as follows: to from quotient images
(SQI) calculated result carries out -0.5 grey scale pixel value offset, then will not be set to 0.5 gray scale in the data of codomain [0,1], finally
The numerical value zoom operations that 255 times are carried out to whole image pixel value, obtain 8bit without symbol shaping grey scale pixel value.
Before treatment in order to accelerate to calculate, need to handle the real-coded GA of sensing results for shape data, this hair
The bright middle shape data for selecting 8bit precision, therefore have following pretreatment:
It is greater than 1 and the data less than 1 in SQI first of all for retaining, the increment of -0.5 gray scale is carried out to whole pixel brightness,
Weaken the amplitude of 0.5 gray scale, filter codomain not in the value of [0,1], be set as 0.5 gray scale:
R (x, y)=R (x, y) -0.5
If (R (x, y) < 0) R (x, y)=0.5
If (R (x, y) > 1) R (x, y)=0.5
Finally zoom in and out processing:
R ' (x, y)=R (x, y) * 255.0
R ' (x, y)=Round (R ' (x, y))
Wherein Round function plays the role of rounding up and switchs to shaping.
Detection method used in step 6 is known as NCC (Normalized Cross Correlation), that is, normalizes
Whether crosscorrelation can will have relationship quantization, often use in big data analysis between data sequence and sequence.About original
Definition, a two-dimensional sequence I1Certain region and another template ItThe degree of correlation can be indicated with following form:
Mean information is respectively as follows:
Complete comparison type is detected, that is, sliding window detection, is had:
Wherein:
In formula, the matrix of (x, y) ∈ [M*N], [m*n] can indicate window size.
Eventually by sliding window mode, a kind of temperature figure based on normalization crosscorrelation property has been obtained.
Matching template comes from pre-training library.
For multiple dirty, temperature figure can become more chaotic, but have no effect on the normal output of self-test, because when not having
When dirty, temperature figure is more clean and tidy and not in threshold range.
Step 8 analyzes connected domain, acquires the position in the connected domain upper left corner, can position digital camera image sensor
Dirty position (as shown in Figure 6).
Claims (7)
1. a kind of dirty self checking method of the digital camera image sensor based on retina perception, it is characterized in that: containing following
Step:
Step 1: placing a pure color object before digital camera camera lens, which is completely covered digital camera camera lens view
Open country is taken pictures with digital camera, obtains the image data of a frame RGB channel；
Step 2: carrying out transcendental method gray processing to the image data；
Step 3: one convolution kernel of setting, it is down-sampled to carry out the mean value based on convolution；
Step 4: obtaining illumination modeling data by gaussian filtering, and using based on view based on the image data that step 3 obtains
The data of the relocatable that sensing results are calculated from quotient images of nethike embrane perception；
Step 5: carrying out data processing, no symbol shape data is obtained；
Step 6: crosscorrelation template matching is normalized using normalization cross correlation algorithm formula, and pass through sliding window
Mouth mode, obtains correlation temperature figure；
Step 7: carrying out thresholding processing to correlation temperature figure；
Step 8: analysis connected domain, obtains the dirty position of digital camera image sensor.
2. the dirty self checking method of the digital camera image sensor according to claim 1 based on retina perception,
It is characterized in: in the transcendental method of the step 2, obtains gray level image using the mode being maximized to RGB channel component, first
Proved recipe method is nonlinear processing function, and the function expression is as follows:
H (x, y)=MAX (R (x, y), G (x, y), B (x, y))
Wherein, H (x, y) is the gray-scale pixel values at position (x, y), MAX (R (x, y), G (x, y), B (x, y)) be ask three it is defeated
Enter the function of maximum value in value, R (x, y), G (x, y), B (x, y) respectively represents RGB channel in the pixel at position (x, y)
Component value.
3. the dirty self checking method of the digital camera image sensor according to claim 1 based on retina perception,
It is characterized in: convolution kernel format in the step 3 are as follows:
Corresponding 3 × 3 convolution kernel calculated results of one pixel, the down-sampled sliding of mean value
Window is with step-length (3,3) for stepping.
4. the dirty self checking method of the digital camera image sensor according to claim 1 based on retina perception,
It is characterized in: conversion calculating of the step 5 using floating-point to unsigned int data, concrete operation step are as follows: to from quotient images
Calculated result carries out -0.5 grey scale pixel value offset, then will not be set to 0.5 gray scale in the data of codomain [0,1], finally to whole
A image pixel value carries out 255 times of numerical value zoom operations and round processing, obtains 8bit without symbol shaping picture
Plain gray value.
5. the dirty self checking method of the digital camera image sensor according to claim 1 based on retina perception,
Be characterized in: matching template comes from pre-training library in the step 6.
6. the dirty self checking method of the digital camera image sensor according to claim 1 based on retina perception,
Be characterized in: pure color object is area source in the step 1.
7. the dirty self checking method of the digital camera image sensor according to claim 6 based on retina perception,
Be characterized in: the area source is planar light source.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910118410.9A CN110012287B (en) | 2019-02-16 | 2019-02-16 | The dirty self checking method of digital camera image sensor based on retina perception |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910118410.9A CN110012287B (en) | 2019-02-16 | 2019-02-16 | The dirty self checking method of digital camera image sensor based on retina perception |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110012287A CN110012287A (en) | 2019-07-12 |
| CN110012287B true CN110012287B (en) | 2019-11-05 |
Family
ID=67165748
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910118410.9A Active CN110012287B (en) | 2019-02-16 | 2019-02-16 | The dirty self checking method of digital camera image sensor based on retina perception |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110012287B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110572636B (en) * | 2019-07-23 | 2020-08-04 | RealMe重庆移动通信有限公司 | Camera contamination detection method and device, storage medium and electronic equipment |
| CN110766672B (en) * | 2019-10-21 | 2020-11-20 | 盐城市钊扬工业设计有限公司 | Electrode fouling detection method |
| CN110559012B (en) * | 2019-10-21 | 2022-09-09 | 江苏鹿得医疗电子股份有限公司 | Electronic stethoscope, control method thereof and control method of medical equipment |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7535501B1 (en) * | 2004-10-12 | 2009-05-19 | Lifetouch, Inc. | Testing of digital cameras |
| WO2009139790A1 (en) * | 2008-05-14 | 2009-11-19 | Nikon Corporation | Device and method for detecting dust on an image sensor |
| CN104867159B (en) * | 2015-06-05 | 2018-04-10 | 北京大恒图像视觉有限公司 | A kind of digital camera sensor dust detection and stage division and device |
| CN107194886B (en) * | 2017-05-03 | 2020-11-10 | 深圳大学 | Dust detection method and device for camera sensor |
-
2019
- 2019-02-16 CN CN201910118410.9A patent/CN110012287B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110012287A (en) | 2019-07-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6997309B2 (en) | Image processing system and processing method using deep neural network | |
| US10592768B2 (en) | Hough processor | |
| CN109583285B (en) | Object recognition method | |
| CN110012287B (en) | The dirty self checking method of digital camera image sensor based on retina perception | |
| CN107209849B (en) | Eye tracking | |
| Kong et al. | A physically-based approach to reflection separation: from physical modeling to constrained optimization | |
| Horprasert et al. | A statistical approach for real-time robust background subtraction and shadow detection | |
| JP6246757B2 (en) | Method and system for representing virtual objects in field of view of real environment | |
| Hui et al. | Reflectance capture using univariate sampling of brdfs | |
| EP1933694B1 (en) | Eye tracker having an extended span of operating distances | |
| Calian et al. | From faces to outdoor light probes | |
| EP3051793B1 (en) | Imaging apparatus, systems and methods | |
| CN101006451A (en) | Method for determining direction of principal light source in image | |
| EP3241151A1 (en) | An image face processing method and apparatus | |
| CN107203743B (en) | Face depth tracking device and implementation method | |
| US11508113B2 (en) | Denoising techniques suitable for recurrent blurs | |
| CN110428394A (en) | Method, apparatus and computer storage medium for target mobile detection | |
| Peng et al. | Improved 3D lighting environment estimation for image forgery detection | |
| KR20230042706A (en) | Neural network analysis of LFA test strips | |
| CN115375991A (en) | Strong/weak illumination and fog environment self-adaptive target detection method | |
| Merkle et al. | State of the art of quality assessment of facial images | |
| Cyganek et al. | Virtual high dynamic range imaging for underwater drone navigation | |
| Zheng et al. | An extended photometric stereo algorithm for recovering specular object shape and its reflectance properties | |
| KR102618580B1 (en) | Nighttime low-light image enhancement method based on retinex and atmospheric light estimation, apparatus and computer program for performing the method | |
| Siriborvornratanakul et al. | Multiscale visual object detection for unsupervised ubiquitous projection based on a portable projector-camera system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20191014 Address after: 518109 zone a, floor 2, building 1, Dabu Lane Industrial Park, Lane 2, Dabu Lane community, Guanlan street, Longhua New District, Shenzhen City, Guangdong Province Applicant after: Shenzhen creation Electronics Co., Ltd. Address before: 450000 South Distribution Building 102, No. 1 Incubation Park, National University Science Park, Zhengzhou High-tech Zone, Henan Province Applicant before: Zhengzhou Chilean Information Technology Co., Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |