CN112384946A - Image dead pixel detection method and device - Google Patents
Image dead pixel detection method and device Download PDFInfo
- Publication number
- CN112384946A CN112384946A CN201880095536.2A CN201880095536A CN112384946A CN 112384946 A CN112384946 A CN 112384946A CN 201880095536 A CN201880095536 A CN 201880095536A CN 112384946 A CN112384946 A CN 112384946A
- Authority
- CN
- China
- Prior art keywords
- pixel
- pixel point
- neural network
- point
- dead
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 46
- 238000013528 artificial neural network Methods 0.000 claims abstract description 169
- 238000000034 method Methods 0.000 claims abstract description 79
- 238000012937 correction Methods 0.000 claims abstract description 36
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000012545 processing Methods 0.000 claims abstract description 27
- 238000013527 convolutional neural network Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 4
- 230000008569 process Effects 0.000 description 28
- 238000013461 design Methods 0.000 description 25
- 238000004891 communication Methods 0.000 description 19
- 210000002569 neuron Anatomy 0.000 description 18
- 230000006870 function Effects 0.000 description 15
- 238000010586 diagram Methods 0.000 description 11
- 238000004422 calculation algorithm Methods 0.000 description 10
- 238000004364 calculation method Methods 0.000 description 7
- 239000011159 matrix material Substances 0.000 description 7
- 230000002093 peripheral effect Effects 0.000 description 5
- 230000007547 defect Effects 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 235000019800 disodium phosphate Nutrition 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000013529 biological neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000003702 image correction Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The application provides an image bad point detection method and device, relates to the field of image processing, and can solve the problems of poor applicability of bad point detection rules and poor detection effect. The method comprises the following steps: selecting pixel points in a preset area around a first pixel point by taking the first pixel point in the image as a center; inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into a neural network for prediction, and acquiring the probability that the first pixel point is a dead pixel; and if the probability is greater than the preset threshold value, determining that the first pixel point is a bad point. The embodiment of the application is used for obtaining the neural network parameters by utilizing the neural network training and then carrying out dead pixel detection and correction by utilizing the neural network and the neural network parameters.
Description
The present application relates to the field of image processing, and in particular, to a method and an apparatus for detecting bad pixels in an image.
Due to the reasons of small sensor area, high pixel density, cost limitation and the like, the camera sensor of the mobile phone has a great number of defects, wherein dead pixels are generated in random events, when a dark scene is exposed for a long time, the probability of the dead pixels can reach 5%, and the quality of an image is seriously influenced. If the correction force is too strong, the loss of details in the picture is easily caused, and false color appears due to the fact that false correction is generated in a high-frequency area; if the correction strength is too weak, the residual of the dead pixel is easy to cause. At present, a dead pixel can be detected by comparing the absolute value, gradient, direction, and threshold value of a pixel, but an experienced person is required to perform repeated debugging based on experience, and the correction parameter is difficult to determine.
For example, taking a pixel point to be detected as a center, taking 5 × 5Bayer pixel points around the pixel point to be detected to establish a matrix, selecting pixel values of pixel points of the same type of channels at different positions around the pixel point to be detected in the matrix according to different color channels for reference, obtaining a pixel difference value or gradient between the pixel point to be detected and each pixel point of the same type of channels, and determining the pixel point to be detected as a dead pixel when the pixel difference value or gradient exceeds a threshold value. When the pixel point to be detected is determined to be a dead pixel, the position of the dead pixel, such as the image edge, can be determined according to the absolute value difference between the pixel point of 8 channels of the same type around the dead pixel, so as to select a corresponding reliable pixel value according to the position of the dead pixel, and assign the reliable pixel value to the current dead pixel for correction.
Whether the pixel point to be detected is a dead pixel or not is determined according to the threshold value, different threshold values need to be set according to different image sensor types or brightness gain values, the specific threshold values need to be finely adjusted according to different brightness gains, the burden of debugging personnel is increased, different detection rules need to be formulated respectively at a high-frequency area and a line pair position due to the diversity of image types, researchers need to set different rules for fitting, the burden of the personnel is increased, and the applicability is poor. Under the condition that the correction force is too strong, pseudo-color is easy to generate, and when the correction force is too weak, the dead pixel detection effect is not ideal enough, and the detection effect is poor.
Disclosure of Invention
The embodiment of the application provides an image dead pixel detection method and device, which can solve the problems of poor applicability of dead pixel detection rules and poor detection effect.
In a first aspect, a method for detecting an image dead pixel is provided, including: selecting pixel points in a preset area around a first pixel point by taking the first pixel point in the image as a center; inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into a neural network for prediction, and acquiring the probability that the first pixel point is a dead pixel; and if the probability is greater than the preset threshold value, determining that the first pixel point is a bad point. Therefore, the mode of combining the neural network with the dead pixel detection is utilized, the problems that in the prior art, the dead pixel detection faces more rules, the debugging is difficult, the calibration is complex, and the applicability of the dead pixel rule is poor can be solved, manual intervention is not needed, and the method has high applicability.
In one possible design, before selecting the pixel points in the preset area around the first pixel point by taking the pixel points as centers, the method further comprises the step of receiving parameters of the neural network, wherein the parameters comprise weight and bias of internal connection of the neural network, so that the first pixel point can be detected in the neural network directly according to the weight and the bias of the internal connection of the neural network, and the probability that the first pixel point is a dead pixel is obtained.
In one possible design, before selecting a pixel point in a preset area around a first pixel point by taking the pixel point as a center, the method further comprises the steps of generating dead pixel samples of different original image samples; and inputting each original picture sample and the corresponding bad sample into a neural network for training, and obtaining the weight and the bias of the internal connection of the neural network, so that the first pixel point can be detected in the neural network according to the weight and the bias of the internal connection of the neural network, and the probability that the first pixel point is a bad point is obtained.
In one possible design, inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into a neural network for prediction, and obtaining the probability that the first pixel point is a dead pixel comprises: inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into a neural network; and calculating according to the pixel value of the first pixel point, the pixel value of the pixel point in the preset area, and the weight and the bias of the internal connection of the neural network, so as to obtain the probability that the first pixel point is a dead pixel. And the pixel points in the preset area are used for assisting in detecting whether the first pixel point is a dead point. The calculation process can be that the product of the pixel value of the pixel point in the neuron in the previous layer and the weight of the internal connection of the neural network is summed to obtain the value of the neuron in the next layer until an output layer is obtained, namely the probability that the first pixel point is a dead pixel, the overall rule for detecting the dead pixel is simple, the difficult calibration under the influence of various rules is avoided, and the method has strong adaptability.
In one possible design, after determining that the first pixel point is a dead pixel, the method further includes correcting the first pixel point according to at least one second pixel point in a preset area and having the same color channel as the first pixel point and the neural network. The method for correcting the dead pixel by using the neural network has high correction precision.
In one possible design, correcting the first pixel point according to at least one second pixel point in the same color channel as the first pixel point in the preset area and the neural network comprises the steps of predicting the probability that the updated first pixel point is a dead point by using the neural network aiming at each second pixel point in the at least one second pixel point, wherein the updated first pixel point is the first pixel point of which the value of the first pixel point is replaced by the value of the second pixel point, and each second pixel point corresponds to a predicted probability; and correcting the first pixel point by using the value of the second pixel point corresponding to the minimum probability in the predicted at least one probability. The correction method is characterized in that pixel points of channels with the same color around the dead pixel are respectively substituted into the dead pixel during dead pixel correction, the confidence coefficient of the non-dead pixel of the dead pixel is input into the neural network to predict, and the value of the maximum confidence coefficient is taken to correct the dead pixel.
In one possible design, the neural network is a fully-connected neural network or a convolutional neural network. When the convolution neural network is adopted for operation, the convolution kernel operation is used, so that the local receptive field is possessed, the local part of the image is sensed, and the local information can be acquired by simulating biological visual connection. And the local information is analyzed through multilayer connection to obtain global information, and the detection is more accurate. And secondly, parameters of the convolutional neural network are shared, so that the number of the parameters can be effectively reduced, and the implementation on hardware with less resources is facilitated.
In a second aspect, a neural network training method is provided, including: generating dead pixel samples of different original picture samples; and inputting each original picture sample and the corresponding bad point sample into a neural network for training, and obtaining parameters of the neural network so as to detect the bad points of the picture according to the parameters.
In one possible design, the method further includes: and sending parameters of the neural network, wherein the parameters comprise weight and bias of internal connection of the neural network, so that equipment receiving the parameters of the neural network can detect dead pixels of the picture, and for a receiving end, training of the parameters of the neural network with complex calculation amount is not needed, and the calculation complexity of the receiving end is reduced.
In one possible design, generating bad samples of different raw picture samples comprises: collecting pictures with different brightness and different scenes as original picture samples; determining small image blocks from each original image sample, wherein the small image blocks are image blocks of a fixed-size area; and modifying the original pixel values of the partial areas of each image block to obtain a corresponding dead pixel sample of each original image sample. Therefore, the method is different from the method for directly collecting the dead pixel sample in the prior art, the dead pixel sample of the original picture sample can be automatically generated, and the collection efficiency of the dead pixel sample is improved.
In a third aspect, an apparatus is provided, comprising: the acquisition unit is used for selecting pixel points in a preset area around a first pixel point by taking the first pixel point in the image as a center; the obtaining unit is further used for inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into the neural network for prediction, and obtaining the probability that the first pixel point is a dead pixel; and the determining unit is used for determining the first pixel point as a dead pixel if the probability is greater than a preset threshold.
In one possible design, the method further includes receiving parameters of the neural network, the parameters including weights and biases of connections inside the neural network.
In one possible design, the obtaining unit is further configured to: generating dead pixel samples of different original picture samples; and inputting each original picture sample and the corresponding bad sample into a neural network for training, and obtaining the weight and the bias of the internal connection of the neural network.
In one possible design, the obtaining unit is configured to: inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into a neural network; and calculating according to the pixel value of the first pixel point, the pixel value of the pixel point in the preset area, and the weight and the bias of the internal connection of the neural network, so as to obtain the probability that the first pixel point is a dead pixel.
In one possible design, further comprising: and the correction unit is used for correcting the first pixel point according to at least one second pixel point and the neural network, wherein the color channel of the second pixel point is the same as that of the first pixel point in the preset area.
In one possible design, the correction unit is configured to: for each second pixel point in at least one second pixel point, predicting the probability that the updated first pixel point is a dead pixel by using a neural network, wherein the updated first pixel point is the first pixel point of which the value of the first pixel point is replaced by the value of the second pixel point, and each second pixel point corresponds to a predicted probability; and correcting the first pixel point by using the value of the second pixel point corresponding to the minimum probability in the predicted at least one probability.
In one possible design, the neural network is a fully-connected neural network or a convolutional neural network.
In a fourth aspect, an apparatus is provided that includes: the generating unit is used for generating dead pixel samples of different original picture samples; and the training unit is used for inputting each original picture sample and the corresponding bad sample into the neural network for training, and acquiring parameters of the neural network so as to detect the picture bad sample according to the parameters.
In one possible design, a sending unit is further included for sending parameters of the neural network, the parameters including weights and biases of connections inside the neural network.
In one possible design, the generation unit is configured to: collecting pictures with different brightness and different scenes as original picture samples; determining small image blocks from each original image sample, wherein the small image blocks are image blocks of a fixed-size area; and modifying the original pixel values of the partial areas of each image block to obtain a corresponding dead pixel sample of each original image sample.
In a fifth aspect, an apparatus is provided that includes a processor and a memory, the memory storing a program, the processor being configured to invoke the program stored in the memory to perform the following process: selecting pixel points in a preset area around a first pixel point by taking the first pixel point in the image as a center; inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into a neural network for prediction, and acquiring the probability that the first pixel point is a dead pixel; and if the probability is greater than the preset threshold value, determining that the first pixel point is a bad point.
In one possible design, the method further includes receiving parameters of the neural network, the parameters including weights and biases of connections within the neural network.
In one possible design, the processor is further to: generating dead pixel samples of different original picture samples; and inputting each original picture sample and the corresponding bad sample into a neural network for training, and obtaining the weight and the bias of the internal connection of the neural network.
In one possible design, the processor is to: inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into a neural network; and calculating according to the pixel value of the first pixel point, the pixel value of the pixel point in the preset area, and the weight and the bias of the internal connection of the neural network, so as to obtain the probability that the first pixel point is a dead pixel.
In one possible design, the processor is further to: and correcting the first pixel point according to at least one second pixel point and the neural network, which are in the same color channel as the first pixel point, in the preset area.
In one possible design, the processor is to: for each second pixel point in at least one second pixel point, predicting the probability that the updated first pixel point is a dead pixel by using a neural network, wherein the updated first pixel point is the first pixel point of which the value of the first pixel point is replaced by the value of the second pixel point, and each second pixel point corresponds to a predicted probability; and correcting the first pixel point by using the value of the second pixel point corresponding to the minimum probability in the predicted at least one probability.
In one possible design, the neural network is a fully-connected neural network or a convolutional neural network.
In a sixth aspect, an apparatus is provided that includes a processor and a memory, the memory storing a program, the processor being configured to invoke the program stored in the memory to perform the following process: generating dead pixel samples of different original picture samples; and inputting each original image sample and the corresponding bad sample into a neural network for training to obtain parameters of the neural network.
In one possible design, a transmitter is further included for transmitting parameters of the neural network, the parameters including weights and biases for connections within the neural network.
In one possible design, the processor is to: collecting pictures with different brightness and different scenes as original picture samples; determining small image blocks from each original image sample, wherein the small image blocks are image blocks of a fixed-size area; and modifying the original pixel values of the partial areas of each image block to obtain a corresponding dead pixel sample of each original image sample.
In a seventh aspect, an embodiment of the present application provides a computer storage medium for storing computer software instructions for the apparatus, which includes a program designed to execute the above aspects.
In an eighth aspect, embodiments of the present application provide a computer program product containing instructions that, when executed on a computer, cause the computer to perform the method of the above aspects.
Fig. 1 is a schematic diagram of a network architecture according to an embodiment of the present application;
fig. 2 is a schematic flowchart of a neural network training method according to an embodiment of the present disclosure;
fig. 3 is a schematic internal structural diagram of a neural network according to an embodiment of the present disclosure;
fig. 4 is a schematic flowchart of a dead pixel detection method according to an embodiment of the present disclosure;
fig. 5 is a schematic flowchart of a dead pixel correction method according to an embodiment of the present application;
fig. 6 is a schematic view of a scenario of an ISP inside a device according to an embodiment of the present application;
fig. 7 is a schematic view of a scenario of an ISP inside a device according to an embodiment of the present application;
fig. 8 is a schematic diagram of an image processing flow according to an embodiment of the present application;
fig. 9 is a schematic structural diagram of a first network device according to an embodiment of the present application;
fig. 10 is a schematic structural diagram of a first network device according to an embodiment of the present application;
fig. 11 is a schematic structural diagram of a first network device according to an embodiment of the present application;
fig. 12 is a schematic structural diagram of a second network device according to an embodiment of the present application.
For ease of understanding, some of the concepts related to the present application are illustratively presented for reference. As follows:
and (3) dead pixel: the array process of each light acquisition point in the image sensor has defects, or errors occur in the process of converting optical signals into digital signals, so that signals of certain pixels obtained by the sensor have errors, and finally, the pixel values of the image are incorrect, and the pixel points with the defects are called as image dead pixels.
Bright spot: the signal value of the pixel point of the image is larger than the normal value, the brightness value is obviously larger than the incident light multiplied by the corresponding proportion, and the brightness of the pixel point can be obviously increased along with the increase of the exposure time.
Dark spot: the signal value of the pixel point of the image is smaller than a normal value, and after the acquisition processing, the signal value is usually 0.
A neural network: also known as Artificial Neural Network (ANN), is an algorithmic mathematical model that mimics animal neural network behavior characteristics to perform distributed parallel information processing. The network achieves the aim of processing information by adjusting the mutual connection relationship among a large number of nodes in the network depending on the complexity of the system.
The method and the device can be used for detecting the image dead pixel, specifically can detect the image dead pixel by using the neural network, and realize the classification detection of complex rules without manual intervention.
The technical solution provided by the present application can be applied to a network architecture as shown in fig. 1, where the network architecture may include a first network device 101 for performing a neural network training procedure and a second network device 102 for performing a picture dead pixel detection and correction procedure. The first network device 101 may be a high-performance computer capable of performing operations on big data, and the second network device 102 may be a mobile terminal, such as a mobile phone, capable of performing functions such as image capturing.
The main process of the application can comprise a neural network training process and an image dead pixel detection and correction process. The two processes can be independent of each other, run on different time and equipment respectively, and can also run on the same equipment, and can be adjusted according to actual conditions. For example, by using the network architecture formed by the computer and the mobile phone, the training process can perform a large amount of image training on a computer with high performance due to the limitation of the terminal computing capability of the mobile phone, so as to obtain neural network parameters, the neural network parameters obtained in the training process are led into the mobile phone from the computer, and the mobile phone can perform dead pixel detection and correction processes on the shot images by using the neural network parameters. Therefore, the bad point detection method and the device have the advantages that the bad point detection face more rules, the debugging is difficult and the like, the neural network and the bad point detection can be combined, the neural network can be used as an end-to-end solution, the neural network can be realized by using codes, the classification detection task of complex rules can be completed without manual intervention, and the applicability is high.
The following is a description of method embodiments of the present application.
The embodiment of the application provides a neural network training method, which comprises the following steps: generating dead pixel samples of different original picture samples; and inputting each original picture sample and the corresponding bad sample into a neural network for training, and obtaining the weight and the bias of the internal connection of the neural network.
It should be noted that the neural network training method may be executed in the first network device, where the first network device sends the trained weights and biases of the internal connections of the neural network to the second network device, so that the second network device performs picture dead pixel detection according to the neural network parameters; or, the neural network training method may also be directly executed in the second network device, and the second network device directly performs the picture dead pixel detection according to the neural network parameters obtained by the second network device.
Whether the neural network training method is performed in the first network device or the second network device, as shown in fig. 2, the method may specifically include:
201. and acquiring pictures with different brightness and different scenes as original picture samples.
In the field of machine learning and cognitive science, a neural network is a mathematical or computational model that mimics the structure and function of a biological neural network, and is used to estimate or approximate functions. Neural networks are computed from a large number of artificial neuron connections. Fig. 3 shows the basic structure of a neural network. The input layer (input layer) may be a pixel value of a pixel point of an image, the hidden layer (hidden layer) is each layer composed of a plurality of neurons and links between the input layer and the output layer, each layer may be a calculation result of weights of a previous layer connected with the neurons, and the output layer (output layer) is an output result formed by information transmission, analysis and weighing in the neuron links. In most cases, the artificial neural network can change the internal structure on the basis of external information, and is an adaptive system. The connections among different neurons have different weights, and the neural network can be better fitted to a practical model through repeated training and adjustment of samples.
In the embodiment of the application, a neural network is used for carrying out sample training on a large number of pictures, neural network parameters used for detecting the dead pixels of the pictures by the neural network are obtained, and the parameters can comprise weight and offset. Various brightnesses can be achieved by adding dead spots in the normal picture, for example bright spots and dark spots can be added. The pixel value of the bright point, namely the image pixel point, is larger than the normal value of the original pixel point, and the pixel value of the dark point, namely the image pixel point, is smaller than the normal value of the original pixel point. Different scenes may for example be pictures taken in different environments, such as different scenes taken in malls, streets and rooms, but may also include pictures taken during the day and night, etc.
202. And determining small image blocks from each original picture sample, wherein the small image blocks are image blocks of a fixed-size area.
For example, the fixed-size region may be a matrix of 7 × 7Bayer size, and the image patch includes 49 pixel points, or may be a matrix of another size, for example, a matrix of 5 × 5Bayer size, and the image patch includes 25 pixel points, which is not limited in this application.
203. And modifying the original pixel values of the partial areas of each image block to obtain a corresponding dead pixel sample of each original image sample.
Specifically, the pixel values of the pixel points in the partial area of each image patch may be modified by a computer, for example, the modified pixel values are greater than the original pixel values, so that the modified pixel points become bright points, or the modified pixel values are smaller than the original pixel values, so that the modified pixel points become dark points, and the image sample with bright points or dark points is a dead point sample.
204. And inputting each original picture sample and the corresponding bad sample into a neural network for training, and obtaining the weight and the bias of the internal connection of the neural network.
Taking fig. 3 as a basic structure of a fully-connected neural network, an input layer is a pixel value of a pixel point of an original picture sample before modification, and each connection between the input layer and a hidden layer, between the hidden layer, and between the hidden layer and an output layer represents a neural network parameter, which includes a weight and an offset. For the neurons 1-6 of the input layer, assigning pixel values of 6 pixel points to the neurons 1-6, multiplying the pixel values of the 6 pixel points by 6 weights respectively, then summing, and adding an offset value to obtain a value of the neuron 7 in the hidden layer. Similarly, when the neurons 1 to 6 are assigned as pixel values of the other 6 pixels, the pixel values of the other 6 pixels are multiplied by 6 weights respectively and then summed, and the offset value is added to obtain the value of the neuron 8 in the hidden layer. In this way, the same calculation method is used between the hidden layers and the output layer, and the weights and offset values between the neurons are modified through multiple iterations, so that the result output by the neuron 14 of the output layer, that is, the probability that the pixel point of the modified dead pixel sample is a dead pixel, is sufficiently high, for example, exceeds the first threshold, and the probability that the pixel point of the original picture sample before modification corresponding to the dead pixel sample is a dead pixel, is sufficiently low, for example, is lower than the second threshold. That is, when the neural network obtains sufficiently high detection precision after training and has the capability of detecting dead pixels, the neural network parameters between the input layer and the hidden layer, between the hidden layer and the hidden layer, and between the hidden layer and the output layer are stored.
After obtaining the neural network parameters obtained by training the neural network, the neural network can be used for detecting dead pixels. In this embodiment, if the neural network training method is executed on a first network device, the neural network parameters trained on the first network device may be imported into a second network device, for example, from a computer to a mobile phone. Therefore, the picture can be detected on line in the mobile phone. Therefore, an embodiment of the present application further provides an image dead pixel detection method, as shown in fig. 4, including:
401. the second network device selects the pixel points in the preset area around the first pixel point by taking the first pixel point in the image as a center.
The second network device may be a mobile phone, and stores the neural network and the neural network parameters imported from the computer. When the mobile phone performs the on-line detection of the actual dead pixel of the image, all the pixel points on the image can be traversed to obtain the dead pixel in the image. For the first pixel point, the first pixel point may be any pixel point in the traversal process. When the first pixel point is detected, the pixel points in the peripheral preset region of the first pixel point may be intercepted with the first pixel point as a center, for example, the peripheral preset region is a 7 × 7 Bayer-sized matrix or a 5 × 5 Bayer-sized matrix with the first pixel point as a center, and the application is not limited. For a pixel point at the edge of an image, a pixel point pair in a preset area around the pixel point can be called as an image patch taking the pixel point at the edge as a center.
402. And the second network equipment inputs the pixel value of the first pixel point and the pixel value of the pixel point in the preset area into the neural network for prediction to obtain the probability that the first pixel point is a dead pixel.
The second network device can input the pixel value of the first pixel point and the pixel value of the pixel point in the preset area into the neural network, and the probability that the first pixel point is a dead pixel is obtained by calculating according to the pixel value of the first pixel point, the pixel value of the pixel point in the preset area, and the weight and the bias of the internal connection of the neural network. And the pixel points in the preset area are used for assisting in predicting whether the first pixel point is a dead pixel in the neural network prediction process.
Exemplarily, after the cell phone inputs the pixel value of the first pixel point and the pixel value of the pixel point in the preset region to the neural network, according to the neural network parameters obtained by training, that is, the weight and the offset between the neurons in the neural network, the pixel value of the first pixel point in the input layer and the weight of the hidden layer are multiplied and summed, and then the offset is added to obtain the pixel value of the neuron of the hidden layer, and then the probability that the first pixel point output by the output layer is a bad point is obtained through the calculation of the pixel value, the weight and the offset of the neuron between the hidden layer and the output layer and the calculation of the pixel value, the weight and the offset of the neuron between the hidden layer and the output layer.
403. And if the probability is greater than the preset threshold value, the second network equipment determines that the first pixel point is a dead point.
For example, the preset threshold may be 85%, 90%, etc., and the application is not limited thereto. And if the probability is greater than the preset threshold value, the first pixel point is considered as a dead pixel and needs to be corrected. The preset threshold may be an empirical value, for example, in order to obtain the empirical value, a picture may be taken through a mobile phone, the step 401 to the step 403 is performed to detect the dead pixel of the image, and it is determined whether a dead pixel remains, if a dead pixel remains, the preset threshold may be adjusted downward, if no dead pixel remains, the preset threshold may be appropriately adjusted upward, and finally a suitable preset threshold is obtained.
Therefore, the neural network is combined with dead pixel detection, the neural network is trained through a large number of diversified images, the neural network is used for carrying out high-level characteristics of image dead pixel online detection, dead pixels are detected from a higher latitude, dead pixel detection has higher accuracy, and the defects of low-level characteristics such as difficulty in selecting threshold values, excessive parameters, difficulty in adjusting and the like are overcome.
If the first pixel point is determined to be a dead pixel, the embodiment of the application further provides a dead pixel correction method, and the second network device can correct the first pixel point according to at least one second pixel point of a channel having the same color as the first pixel point in the preset area and the neural network. As shown in fig. 5, the correction method may specifically include:
501. the second network equipment predicts the probability that the updated first pixel point is a dead pixel by using a neural network aiming at each second pixel point in at least one second pixel point, the updated first pixel point is the first pixel point after the value of the first pixel point is replaced by the value of the second pixel point, and each second pixel point corresponds to a predicted probability.
The first pixel point is a central point in a preset area. The color channel may be a channel of three colors of R (red), G (green) and B (blue), and the pixels of the same color have the same color channel. In order to obtain the real pixel value of the dead pixel, the pixel values of the same color channels around the dead pixel in the preset area can be extracted, and are sequentially and respectively substituted into the dead pixel position to be input into the neural network for detection, so that the probability that the pixel point at the dead pixel position is the dead pixel after updating is predicted.
When the first pixel point is a dead pixel, the pixel values of the pixel points of the same color channel around the first pixel point are closest to the pixel value of the first pixel point, so that the pixel points of the same color channel around the first pixel point can be extracted to obtain at least one second pixel point, the pixel value of the first pixel point is updated according to the pixel value of the first second pixel point in the at least one second pixel point, the updated pixel value of the first pixel point is input into a neural network for prediction, and the updated probability that the first pixel point is a dead pixel is obtained. And then the second network equipment continuously updates the pixel value of the first pixel point by the pixel value of the next second pixel point and inputs the pixel value of the first pixel point into the neural network for prediction, and the probability that the updated first pixel point is a dead pixel is obtained until the predicted probability corresponding to all the at least one second pixel point is obtained.
502. And the second network equipment corrects the first pixel point by using the value of the second pixel point corresponding to the minimum probability in the predicted at least one probability.
In another words, after predicting, by the second network device, the probability that the updated first pixel is a dead pixel for each second pixel in the at least one second pixel, the second network device takes the pixel value of the second pixel with the highest confidence coefficient to modify the pixel value of the first pixel, thereby completing dead pixel correction.
That is, when the mobile phone performs image correction, the pixel values of the pixel points of the same channel around the dead pixel are extracted and respectively substituted into the dead pixel positions to be input into the neural network for detection, so as to obtain the prediction probability of no dead pixel after the dead pixel is corrected, and the pixel value with the highest confidence coefficient is taken as the real pixel value of the dead pixel to be modified, so that the dead pixel correction is completed. Therefore, the confidence coefficient of the peripheral same-channel pixel values is predicted by utilizing the neural network, the precision of dead pixel correction can be improved, and the correction logic is simple.
In the detection and correction process, when the mobile phone processes the image, dead pixel correction can be performed in the Bayer domain. The dead pixel detection and correction algorithm process can be selected to run on different operation platforms according to actual needs. The platforms capable of running the algorithm include a hardware algorithm running on an Image Signal Processor (ISP) platform and a software algorithm running on a Central Processing Unit (CPU) platform. When the algorithm is integrated into an ISP hardware platform, the operation speed is higher, the power consumption is lower, the requirement of high-definition image real-time processing can be met, but the cost is higher, the iteration speed is lower, and a functional circuit is difficult to multiplex; when the algorithm is integrated by a software scheme and is processed by a CPU, multiple software iterations are conveniently performed, the cost is low, the operation speed is low, the power consumption is high, and different algorithm operation platforms can be selected according to actual conditions when the algorithm is selected.
For a scene in which a hardware algorithm runs on an ISP platform, the ISP may be integrated on an Application Processor (AP), as shown in fig. 6, after an image is obtained by shooting through a camera module, the image is processed through the AP, including dead pixel detection and correction of the image through the ISP platform in the AP. The ISP may also be an independent chip, as shown in fig. 7, after the image is obtained by shooting through the camera module, the image is first detected and corrected for dead pixels through the ISP platform, and then the detected and corrected image is further processed through the AP.
On the ISP platform, the dead pixel is corrected as a module in the image processing flow. Image processing includes black level correction (to remove bottom current noise), dead-point detection and correction (to remove dead-point data in the sensor), de-noising (to remove noise), conversion of raw data to RGB data (Demosaic), 3A (for auto-white balance, auto-focus and auto-exposure), sharpening (to adjust sharpness), color space conversion (for conversion to a different color space for processing), and color enhancement (optional, to adjust color). After the processing, the picture is converted from an original digital signal into an image which meets the visual requirement of human eyes. The position of the dead pixel correction in the image processing is shown in fig. 8.
It should be noted that, the method for correcting the dead pixel through the neural network in the present application does not generate misjudgment on the high frequency region, can effectively avoid the generation of pseudo color, and does not generate influence on the detail texture of the image.
The neural network involved in the embodiment of the application can be a fully-connected neural network, and a deep neural network can be adopted for on-line dead pixel detection and correction. At present, the neural network structure includes a convolutional neural network LeNet5, a deep convolutional neural network structure AlexNet, a Visual Geometry Group (VGG) 16, a residual error network ResNet, a convolutional neural network GoogLeNet with an inclusion structure, and the like, and different neural networks all have dead pixel detection capability and can better perform dead pixel detection and correction tasks. According to the method and the device, a convolutional neural network can be adopted to detect and correct the dead pixel according to actual operation capacity.
The method is characterized in that a convolution kernel operation is used, so that the method has a local receptive field, senses the local part of an image, and simulates biological visual nerve connection to acquire local information. And through multilayer connection, the local information is analyzed to obtain global information, the detection is more accurate, and the precision of the dead pixel detection by using the convolutional neural network can be improved by about 5 percent compared with the precision of the fully-connected neural network. And secondly, parameters of the convolutional neural network are shared, so that the number of the parameters can be effectively reduced, and the implementation on hardware with less resources is facilitated.
The method and the device can be applied to mobile photographing equipment for carrying out dead pixel online detection, can also be applied to hole or dead pixel correction of the depth picture, and adopt dead pixel correction to avoid dead pixels of display equipment when Augmented Reality (AR) or Virtual Reality (VR) displays. In addition, the method can also be applied to equipment such as the Internet of vehicles and the Internet of things.
The above-mentioned scheme provided by the embodiment of the present application is introduced mainly from the perspective of interaction between network elements. It is to be understood that each network element, for example, the first network device, the second network device, etc., contains corresponding hardware structures and/or software modules for performing each function in order to realize the functions. Those of skill in the art would readily appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as hardware or combinations of hardware and computer software. Whether a function is performed as hardware or computer software drives hardware depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
In the embodiment of the present application, the first network device, the second network device, and the like may be divided into functional modules according to the above method example, for example, each functional module may be divided corresponding to each function, or two or more functions may be integrated into one processing module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. It should be noted that, in the embodiment of the present application, the division of the module is schematic, and is only one logic function division, and there may be another division manner in actual implementation.
In the case of dividing each functional module by corresponding functions, fig. 9 shows a possible structural schematic diagram of the first network device involved in the foregoing embodiment, and the first network device 90 includes: an acquisition unit 901 and an acquisition unit 902. The acquisition unit 901 is configured to support the first network device to execute the process 201 in fig. 2, and the acquisition unit 902 is configured to support the first network device to execute the processes 202, 203, and 204 in fig. 2. All relevant contents of each step related to the above method embodiment may be referred to the functional description of the corresponding functional module, and are not described herein again.
Fig. 10 shows a schematic diagram of a possible structure of the first network device involved in the above-described embodiment, in the case of an integrated unit. The first network device 100 includes: a processing module 1002 and a communication module 1003. Processing module 1302 is configured to control and manage actions of the first network device, e.g., processing module 1002 is configured to enable the first network device to perform processes 201, 202, 203, and 204 in fig. 4, and/or other processes for the techniques described herein. The communication module 1003 is configured to support communication between the first network device and other network entities, for example, the network entities shown in fig. 1. The first network device may further comprise a storage module 1001 for storing program codes and data of the first network device.
The Processing module 1002 may be a Processor or a controller, such as a Central Processing Unit (CPU), a general-purpose Processor, a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, a transistor logic device, a hardware component, or any combination thereof. Which may implement or perform the various illustrative logical blocks, modules, and circuits described in connection with the disclosure. The processor may also be a combination of computing functions, e.g., comprising one or more microprocessors, DSPs, and microprocessors, among others. The communication module 1003 may be a transceiver, a transceiver circuit or a communication interface, etc. The storage module 1001 may be a memory.
When the processing module 1002 is a processor, the communication module 1003 is a communication interface, and the storage module 1001 is a memory, the first network device according to the embodiment of the present application may be the first network device shown in fig. 11.
Referring to fig. 11, the first network device 110 includes: a processor 1102, a communication interface 1103, a memory 1101, and a bus 1104. Wherein the transceiver communication interface 1103, the processor 1102 and the memory 1101 are connected to each other by a bus 1104; the bus 1104 may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown in FIG. 11, but this is not intended to represent only one bus or type of bus.
In the case of dividing each functional module by corresponding functions, fig. 12 shows a schematic diagram of a possible structure of the second network device involved in the foregoing embodiments, and the second network device 120 includes: an acquisition unit 1201, a determination unit 1202, a correction unit 1203, and a reception unit 1204. The obtaining unit 1201 is configured to support the second network device to perform the processes 401 and 402 in fig. 4, the processes 502 and 503 in fig. 5, the determining unit 1202 is configured to support the second network device to perform the process 403 in fig. 4, the correcting unit 1203 is configured to support the second network device to perform the processes 501 and 502 in fig. 5, and the receiving unit 1204 may be configured to receive the neural network parameters from the first network device. All relevant contents of each step related to the above method embodiment may be referred to the functional description of the corresponding functional module, and are not described herein again.
In the case of using an integrated unit, a schematic diagram of a possible structure of the second network device in the above embodiment may refer to fig. 10, and as the structure of the first network device, may include: a processing module and a communication module. The processing module is used to control and manage the actions of the second network device, e.g., the processing module is used to support the second network device to perform processes 401, 402, and 403 in fig. 4, processes 501 and 502 in fig. 5, and/or other processes for the techniques described herein. The communication module is configured to support communication between the second network device and other network entities, for example, the network entity shown in fig. 1. The second network device may also include a storage module to store program code and data for the second network device.
The processing module in the second network device may be a processor or a controller, and may be, for example, a central processing unit CPU, a general purpose processor, a digital signal processor DSP, an application specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic device, a transistor logic device, a hardware component, or any combination thereof. Which may implement or perform the various illustrative logical blocks, modules, and circuits described in connection with the disclosure. The processor may also be a combination of computing functions, e.g., comprising one or more microprocessors, DSPs, and microprocessors, among others. The communication module in the second network device may be a transceiver, a transceiving circuit or a communication interface, etc. The storage module in the second network device may be a memory.
Similar to the structure of the first network device in fig. 11, when the processing module in the second network device is a processor, the communication module is a communication interface, and the storage module is a memory, the second network device may include: a processor, a communication interface, a memory, and a bus. The communication interface, the processor and the memory are connected with each other through a bus; the bus may be a peripheral component interconnect standard PCI bus or an extended industry standard architecture EISA bus or the like. The bus may be divided into an address bus, a data bus, a control bus, etc.
The steps of a method or algorithm described in connection with the disclosure herein may be embodied in hardware or in software instructions executed by a processor. The software instructions may be comprised of corresponding software modules that may be stored in Random Access Memory (RAM), flash Memory, Read Only Memory (ROM), Erasable Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), registers, a hard disk, a removable disk, a compact disc Read Only Memory (CD-ROM), or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. Of course, the storage medium may also be integral to the processor. The processor and the storage medium may reside in an ASIC. Additionally, the ASIC may reside in a core network interface device. Of course, the processor and the storage medium may reside as discrete components in a core network interface device.
Those skilled in the art will recognize that in one or more of the examples described above, the functions described herein may be implemented in hardware, software, firmware, or any combination thereof. When implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a general purpose or special purpose computer.
The above description is only an embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions within the technical scope of the present invention are intended to be covered by the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.
Claims (23)
- An image dead pixel detection method is characterized by comprising the following steps:selecting pixel points in a preset area around a first pixel point by taking the first pixel point in an image as a center;inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into a neural network for prediction to obtain the probability that the first pixel point is a dead pixel;and if the probability is greater than a preset threshold value, determining that the first pixel point is a dead point.
- The method of claim 1, wherein before selecting the pixels in the predetermined area around the first pixel with the pixel as a center, the method further comprises:receiving parameters of the neural network, the parameters including weights and biases of connections within the neural network.
- The method of claim 1, wherein before selecting the pixels in the predetermined area around the first pixel with the pixel as a center, the method further comprises:generating dead pixel samples of different original picture samples;and inputting each original picture sample and the corresponding bad sample into the neural network for training, and obtaining the weight and the bias of the internal connection of the neural network.
- The method of claim 2 or 3, wherein the step of inputting the pixel value of the first pixel and the pixel values of the pixels in the preset region into a neural network for prediction, and the step of obtaining the probability that the first pixel is a dead pixel comprises:inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into the neural network;and calculating according to the pixel value of the first pixel point, the pixel value of the pixel point in the preset area, the weight connected inside the neural network and the bias, and acquiring the probability that the first pixel point is the dead pixel.
- The method of any of claims 1-4, wherein after determining that the first pixel is a dead pixel, the method further comprises:and correcting the first pixel point according to at least one second pixel point in the preset region and the same color channel as the first pixel point and the neural network.
- The method of claim 5, wherein the correcting the first pixel according to at least one second pixel in the same color channel as the first pixel in the preset region and the neural network comprises:for each second pixel point in the at least one second pixel point, predicting the probability that the updated first pixel point is a dead pixel by using the neural network, wherein the updated first pixel point is the first pixel point of which the value of the first pixel point is replaced by the value of the second pixel point, and each second pixel point corresponds to a predicted probability;and correcting the first pixel point by using the value of the second pixel point corresponding to the minimum probability in the predicted at least one probability.
- The image dead pixel detection method according to any one of claims 1 to 6, wherein the neural network is a fully connected neural network or a convolutional neural network.
- An apparatus, comprising:the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for selecting pixel points in a preset area around a first pixel point by taking the first pixel point in an image as a center;the obtaining unit is further configured to input the pixel value of the first pixel point and the pixel values of the pixel points in the preset region into a neural network for prediction, so as to obtain a probability that the first pixel point is a dead pixel;and the determining unit is used for determining the first pixel point as a dead pixel if the probability is greater than a preset threshold.
- The apparatus of claim 8, further comprising a receiving unit configured to receive parameters of the neural network, the parameters including weights and biases of connections within the neural network.
- The apparatus of claim 8, wherein the obtaining unit is further configured to:generating dead pixel samples of different original picture samples;and inputting each original picture sample and the corresponding bad sample into the neural network for training, and obtaining the weight and the bias of the internal connection of the neural network.
- The apparatus according to claim 9 or 10, wherein the obtaining unit is configured to:inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into the neural network;and calculating according to the pixel value of the first pixel point, the pixel value of the pixel point in the preset area, the weight connected inside the neural network and the bias, and acquiring the probability that the first pixel point is the dead pixel.
- The apparatus according to any one of claims 8-11, further comprising a calibration unit for calibrating the first pixel point according to at least one second pixel point in the same color channel as the first pixel point in the preset region and the neural network.
- The apparatus of claim 12, wherein the correction unit is configured to:for each second pixel point in the at least one second pixel point, predicting the probability that the updated first pixel point is a dead pixel by using the neural network, wherein the updated first pixel point is the first pixel point of which the value of the first pixel point is replaced by the value of the second pixel point, and each second pixel point corresponds to a predicted probability;and correcting the first pixel point by using the value of the second pixel point corresponding to the minimum probability in the predicted at least one probability.
- The apparatus of any one of claims 8-13, wherein the neural network is a fully-connected neural network or a convolutional neural network.
- An apparatus comprising a processor and a memory, the memory for storing a program, the processor for invoking the program stored in the memory to perform the following:selecting pixel points in a preset area around a first pixel point by taking the first pixel point in an image as a center; inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into a neural network for prediction to obtain the probability that the first pixel point is a dead pixel; and if the probability is greater than a preset threshold value, determining that the first pixel point is a dead point.
- The apparatus of claim 15, further comprising a receiver for receiving parameters of the neural network, the parameters including weights and biases for connections within the neural network.
- The apparatus of claim 15, wherein the processor is further configured to:generating dead pixel samples of different original picture samples;and inputting each original picture sample and the corresponding bad sample into the neural network for training, and obtaining the weight and the bias of the internal connection of the neural network.
- The apparatus of claim 16 or 17, wherein the processor is configured to:inputting the pixel value of the first pixel point and the pixel values of the pixel points in the preset area into the neural network;and calculating according to the pixel value of the first pixel point, the pixel value of the pixel point in the preset area, the weight connected inside the neural network and the bias, and acquiring the probability that the first pixel point is the dead pixel.
- The apparatus according to any of claims 15-18, wherein the processor is further configured to: and correcting the first pixel point according to at least one second pixel point in the preset region and the same color channel as the first pixel point and the neural network.
- The apparatus of claim 19, wherein the processor is configured to:for each second pixel point in the at least one second pixel point, predicting the probability that the updated first pixel point is a dead pixel by using the neural network, wherein the updated first pixel point is the first pixel point of which the value of the first pixel point is replaced by the value of the second pixel point, and each second pixel point corresponds to a predicted probability;and correcting the first pixel point by using the value of the second pixel point corresponding to the minimum probability in the predicted at least one probability.
- The apparatus of any one of claims 15-20, wherein the neural network is a fully-connected neural network or a convolutional neural network.
- A computer storage medium for storing computer software instructions for use by the apparatus, comprising instructions for executing a program as claimed in claims 1 to 7.
- A computer program product comprising instructions, characterized in that when run on a computer, the computer is adapted to execute the instructions as designed in the preceding claims 1 to 7.
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2018/095712 WO2020010638A1 (en) | 2018-07-13 | 2018-07-13 | Method and device for detecting defective pixel in image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112384946A true CN112384946A (en) | 2021-02-19 |
Family
ID=69143145
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201880095536.2A Pending CN112384946A (en) | 2018-07-13 | 2018-07-13 | Image dead pixel detection method and device |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN112384946A (en) |
| WO (1) | WO2020010638A1 (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113657238A (en) * | 2021-08-11 | 2021-11-16 | 南京精益安防系统科技有限公司 | Fire early warning method based on neural network, storage medium and terminal equipment |
| CN114677574A (en) * | 2022-05-26 | 2022-06-28 | 杭州宏景智驾科技有限公司 | Method and system for diagnosing image fault for automatic driving |
| CN116228756A (en) * | 2023-05-08 | 2023-06-06 | 常州星宇车灯股份有限公司 | Method and system for detecting bad points of camera in automatic driving |
| CN116886894A (en) * | 2023-08-04 | 2023-10-13 | 上海宇勘科技有限公司 | Picture dead pixel detection and correction method for adjacent internal column dead pixels |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20220004436A (en) * | 2020-07-03 | 2022-01-11 | 삼성전자주식회사 | An image processing apparatus including a neural network processor and operating method thereof |
| EP4241438A1 (en) * | 2022-01-12 | 2023-09-13 | Dream CHIP Technologies GmbH | Method and image processor unit for detecting defective pixels |
| CN114494260B (en) * | 2022-04-18 | 2022-07-19 | 深圳思谋信息科技有限公司 | Object defect detection method and device, computer equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101715050A (en) * | 2009-11-20 | 2010-05-26 | 北京中星微电子有限公司 | Method and system for detecting dead pixels of image sensor |
| CN101807245A (en) * | 2010-03-02 | 2010-08-18 | 天津大学 | Artificial neural network-based multi-source gait feature extraction and identification method |
| CN103475828A (en) * | 2013-10-10 | 2013-12-25 | 旗瀚科技有限公司 | Method for rectifying missing pixels and image sensor |
| CN104715282A (en) * | 2015-02-13 | 2015-06-17 | 浙江工业大学 | Data prediction method based on improved PSO-BP neural network |
| CN105868797A (en) * | 2015-01-22 | 2016-08-17 | 深圳市腾讯计算机系统有限公司 | Network parameter training method, scene type identification method and devices |
| US20160337971A1 (en) * | 2010-06-04 | 2016-11-17 | Board Of Regents, The University Of Texas System | Wireless communication methods, systems, and computer program products |
| CN106210712A (en) * | 2016-08-11 | 2016-12-07 | 上海大学 | A kind of dead pixel points of images detection and processing method |
| CN107316295A (en) * | 2017-07-02 | 2017-11-03 | 苏州大学 | A kind of fabric defects detection method based on deep neural network |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0676069A (en) * | 1992-08-27 | 1994-03-18 | Toyota Central Res & Dev Lab Inc | Surface defect detecting device |
| JPH06119454A (en) * | 1992-10-08 | 1994-04-28 | Babcock Hitachi Kk | Method and device for detecting abnormality |
| CN106845556A (en) * | 2017-02-09 | 2017-06-13 | 东华大学 | A kind of fabric defect detection method based on convolutional neural networks |
| CN107169956B (en) * | 2017-04-28 | 2020-02-14 | 西安工程大学 | Color woven fabric defect detection method based on convolutional neural network |
-
2018
- 2018-07-13 WO PCT/CN2018/095712 patent/WO2020010638A1/en active Application Filing
- 2018-07-13 CN CN201880095536.2A patent/CN112384946A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101715050A (en) * | 2009-11-20 | 2010-05-26 | 北京中星微电子有限公司 | Method and system for detecting dead pixels of image sensor |
| CN101807245A (en) * | 2010-03-02 | 2010-08-18 | 天津大学 | Artificial neural network-based multi-source gait feature extraction and identification method |
| US20160337971A1 (en) * | 2010-06-04 | 2016-11-17 | Board Of Regents, The University Of Texas System | Wireless communication methods, systems, and computer program products |
| CN103475828A (en) * | 2013-10-10 | 2013-12-25 | 旗瀚科技有限公司 | Method for rectifying missing pixels and image sensor |
| CN105868797A (en) * | 2015-01-22 | 2016-08-17 | 深圳市腾讯计算机系统有限公司 | Network parameter training method, scene type identification method and devices |
| CN104715282A (en) * | 2015-02-13 | 2015-06-17 | 浙江工业大学 | Data prediction method based on improved PSO-BP neural network |
| CN106210712A (en) * | 2016-08-11 | 2016-12-07 | 上海大学 | A kind of dead pixel points of images detection and processing method |
| CN107316295A (en) * | 2017-07-02 | 2017-11-03 | 苏州大学 | A kind of fabric defects detection method based on deep neural network |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113657238A (en) * | 2021-08-11 | 2021-11-16 | 南京精益安防系统科技有限公司 | Fire early warning method based on neural network, storage medium and terminal equipment |
| CN113657238B (en) * | 2021-08-11 | 2024-02-02 | 南京精益安防系统科技有限公司 | Fire early warning method based on neural network, storage medium and terminal equipment |
| CN114677574A (en) * | 2022-05-26 | 2022-06-28 | 杭州宏景智驾科技有限公司 | Method and system for diagnosing image fault for automatic driving |
| CN116228756A (en) * | 2023-05-08 | 2023-06-06 | 常州星宇车灯股份有限公司 | Method and system for detecting bad points of camera in automatic driving |
| CN116886894A (en) * | 2023-08-04 | 2023-10-13 | 上海宇勘科技有限公司 | Picture dead pixel detection and correction method for adjacent internal column dead pixels |
| CN116886894B (en) * | 2023-08-04 | 2024-03-15 | 上海宇勘科技有限公司 | Picture dead pixel detection and correction method for adjacent internal column dead pixels |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020010638A1 (en) | 2020-01-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112384946A (en) | Image dead pixel detection method and device | |
| CN111741211B (en) | Image display method and apparatus | |
| CN101527860B (en) | White balance control apparatus, control method therefor, and image sensing apparatus | |
| US9432589B2 (en) | Systems and methods for generating high dynamic range images | |
| CN111292264A (en) | Image high dynamic range reconstruction method based on deep learning | |
| CN114972085B (en) | Fine granularity noise estimation method and system based on contrast learning | |
| CN112840636A (en) | Image processing method and device | |
| CN112132172A (en) | Model training method, device, equipment and medium based on image processing | |
| CN116091900A (en) | Image processing method, device and storage medium | |
| CN113962859B (en) | Panorama generation method, device, equipment and medium | |
| CN113538223A (en) | Noise image generation method, noise image generation device, electronic device, and storage medium | |
| US11443414B2 (en) | Image signal processing | |
| CN114125280A (en) | Camera exposure control method, device, equipment, storage medium and program product | |
| CN113888509A (en) | Method, device and equipment for evaluating image definition and storage medium | |
| CN111553474A (en) | Ship detection model training method and ship tracking method based on unmanned aerial vehicle video | |
| US20230267708A1 (en) | Method for generating a learning model, a program, and an information processing apparatus | |
| CN115426487A (en) | Color correction matrix adjusting method and device, electronic equipment and readable storage medium | |
| US9210386B2 (en) | Filter setup learning for binary sensor | |
| CN114882385A (en) | Method for counting wheat ears in field based on unmanned aerial vehicle platform | |
| CN113240759A (en) | Visibility determination method, device, equipment and storage medium based on color temperature | |
| CN114255193A (en) | Board card image enhancement method, device, equipment and readable storage medium | |
| CN112734673A (en) | Low-illumination image enhancement method and system based on multi-expression fusion | |
| JP2021093694A (en) | Information processing apparatus and method for controlling the same | |
| WO2023240651A1 (en) | Image processing method and apparatus | |
| US20230088317A1 (en) | Information processing apparatus, information processing method, and storage medium |
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 |