CN115861297A - Printing plate dot image detection and segmentation method and device based on deep learning - Google Patents
Printing plate dot image detection and segmentation method and device based on deep learning Download PDFInfo
- Publication number
- CN115861297A CN115861297A CN202310109402.4A CN202310109402A CN115861297A CN 115861297 A CN115861297 A CN 115861297A CN 202310109402 A CN202310109402 A CN 202310109402A CN 115861297 A CN115861297 A CN 115861297A
- Authority
- CN
- China
- Prior art keywords
- printing plate
- image
- edge
- plate dot
- deep learning
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
A method and a device for detecting and segmenting printing plate dot images based on deep learning are provided, and a novel method, a novel device and a novel device for detecting and segmenting the printing plate dot images are provided by utilizing a deep learning technology aiming at the problem of calculating the printing plate dot area rate quality index in printing quality control. The traditional printing plate dot image detection method based on threshold segmentation is sensitive to noise, and difficult in edge positioning and fine edge processing, so that the printing plate dot image segmentation effect is poor. The invention utilizes the deep learning technology to apply the convolutional neural network to the printing plate dot image detection, so that the efficiency and the precision of the printing plate dot image edge detection are greatly improved, and a more accurate calculation result of the printing plate dot area rate can be obtained.
Description
Technical Field
The invention relates to a printing plate dot image detection and segmentation method and device based on deep learning.
Background
The printed matter consists of a printing stock and ink, wherein the ink is regularly distributed on the printing stock in a dot form and presents different spectral colors through superposition of multicolor ink. The production process of printing is a process of transferring ink to a printing stock according to a preset pattern of a printing plate and can be divided into two links of a prepress process and an actual printing process. The prepress process mainly completes the printing plate manufacturing, including image processing, color separation and screening, film printing, plate making, proofing and the like; the actual printing process utilizes the manufactured printing plate to print in the printing machine. Thus, quality control of the printing process is mainly achieved by adjusting for a two-stage process of pre-press and actual printing.
In the main printing methods such as offset printing and gravure printing, patterns are composed of dots, and the dots are the smallest copy units of printing. According to different screening principles, printing screen points can be divided into frequency modulation screen points, amplitude modulation screen points and mixed screening screen points; according to the difference of the shapes, the patterns can be divided into dots such as circles, ellipses, rhombuses and the like. Dot quality is one of the important factors affecting the quality of printed matter during printing. The process of controlling the quality of the printed matter is the process of controlling the accurate transfer and copying of the printing dots, and the guarantee of the quality of the printing dots has very important significance for printing production.
In order to accurately detect and analyze the quality of the printed matter, printing plates with different shapes and different screening types and the dots of the printed matter need to be accurately detected and divided, and the dot area rate or dot coverage rate of the printing plates and the printed matter, namely the dot proportion in unit area, is calculated. The area ratio after the original is subjected to color separation and screening, namely the screening area ratio; the area rates of the printing plate and the printed matter, namely the area rate of printing plate dots and the area rate of printed matter dots, are compared, and the printing quality can be evaluated and identified. In a word, the dot image of each monochromatic ink is obtained from the color printing image, and the quality indexes such as the dot area rate of the printing plate and the printed matter are calculated at the same time, so that the core of the printing quality control is realized; the dot detection and segmentation of printing plates and prints are key links in the control of printing quality.
At present, the detection and segmentation of printing plate dot images need to utilize a camera to collect the printing plate dot images, and then the dot images are transmitted to a computer to be extracted by utilizing an edge detection or image segmentation method. However, in the process of transfer and copying, the printing plate dots are affected by factors such as dot enlargement, printing pressure variation, ink flow and the like, and are interfered by factors such as light scattering, image noise and the like, so that the dot images are prone to edge blurring, and it is difficult to accurately detect the edges of the printing plate dot images by using an optical instrument. Therefore, a number of methods for detecting and segmenting plate dot images have been proposed.
The invention CN112258501A provides a printing plate dot detection and dot area rate calculation method. The specific content comprises the following steps: obtaining a printing plate dot image to be detected, and carrying out image segmentation on the printing plate dot image based on a threshold value to obtain an image segmentation result; carrying out image binarization based on the obtained printing plate dot image segmentation result to obtain a binary image; and carrying out dot detection on the printing plate dot image based on the obtained binary image to obtain a target dot detection result. The method for detecting the printing plate dot image adopts a threshold value method to segment the printing plate image to obtain the segmented result of the printing plate dot image. However, the image segmentation method based on the threshold is sensitive to noise, has low robustness, is difficult to obtain a proper threshold for printing plate images with insignificant gray level difference and overlapped different target gray levels, and has poor segmentation effect.
The invention CN109727232A provides a printing plate net point detection and net point area rate calculation method. The specific content comprises the following steps: collecting a dot image of the printing plate; carrying out graying processing on the dot image to generate a grayscale image; binarizing the gray level image by utilizing an Otsu method to generate a binary image; filtering the binary image from left to right and from top to bottom to generate a filtered image; and calculating the ratio of the number of pixels representing the dots in the filtered image to the total number of pixels of the filtered image to obtain the dot area rate of the printing plate. The printing plate network image detection method described by the invention adopts the Otsu method to carry out binaryzation processing on the printing plate gray level image to obtain a printing plate binaryzation image; and filtering the obtained binary image in a sequence from left to right and from top to bottom to obtain a filtered printing plate dot image segmentation result. Because the Otsu method is also an image segmentation method based on an image global threshold, the gray distribution of an image is taken as the basis for segmenting the image, and the image is quite sensitive to noise; when the size ratio of the screen point target and the background of the printing plate image is very different and the variance function between classes may present double peaks or multiple peaks, the segmentation effect is not good.
Disclosure of Invention
The invention provides a printing plate dot image detection and segmentation method based on deep learning, and provides a novel printing plate dot image detection and segmentation method, device and equipment by utilizing a deep learning technology aiming at the calculation problem of printing plate dot area rate quality indexes in printing quality control. The traditional printing plate dot image detection method based on threshold segmentation is sensitive to noise, and difficult in edge positioning and fine edge processing, so that the printing plate dot image segmentation effect is poor. The invention utilizes deep learning technology to apply Convolutional Neural Network (CNN) to the printing plate dot image detection, so that the efficiency and the precision of the printing plate dot image edge detection are greatly improved, and a more accurate calculation result of the printing plate dot area rate can be obtained.
A printing plate dot image detection and segmentation method based on deep learning comprises the following steps:
step one, labeling image data: drawing a certain number of edge image samples aiming at the acquired high-resolution printing plate dot images;
step two, constructing a network model: a network model is constructed by utilizing a convolutional neural network, and rich hierarchical representation is automatically learned; simultaneously designing a loss function, calculating the difference between a predicted value and an actual value, and using the difference to measure the result of model prediction;
step three, training a network model: training the network model constructed in the second step by using the image data marked in the first step until the loss function is converged to obtain an end-to-end deep learning edge detection model for the printing plate mesh point image;
step four, detecting a new image: using the trained network model in the step three to carry out edge detection on the printing plate dot image to be detected to obtain an edge probability graph of the printing plate dot image;
step five, correcting the detection result: refining the edge detection result of the edge probability graph obtained in the fourth step by using a non-maximum inhibition method; and setting a threshold value by a data set scale optimization method and carrying out binarization processing, thereby obtaining a final detection and segmentation result of the printing plate mesh point image to be detected.
Preferably, the labeling of image data in the first step of the present invention specifically comprises: carrying out artificial edge marking on the acquired high-resolution printing plate dot image; meanwhile, a data enhancement method is used for rotating, mirroring and shearing the dot images, and the scale of the data set is enlarged.
Preferably, the convolutional neural network infrastructure in step two of the present invention includes: at least one of LeNet5, alexNet, VGGNet, *** Inception Net and ResNet.
Preferably, the method for training the network model in step three of the present invention includes at least one of a stochastic gradient descent method SGD, a Momentum gradient descent Momentum, an optimizer RMSprop and Adam.
A printing plate dot image detection and segmentation device based on deep learning comprises:
the image data labeling module is used for drawing a certain number of edge image samples aiming at the acquired high-resolution printing plate dot images;
the network model building module is used for building a network model by utilizing a convolutional neural network and automatically learning rich hierarchical representation; simultaneously designing a loss function, and calculating the difference between a predicted value and an actual value so as to measure the result of model prediction;
the network model training module is used for training the constructed network model by using the marked image data until the loss function is converged to obtain an end-to-end deep learning edge detection model aiming at the printing plate dot image;
the image detection module is used for carrying out edge detection on the printing plate dot image to be detected by using the trained network model to obtain an edge probability graph of the printing plate dot image;
the detection result correction module is used for refining the edge detection result of the obtained edge probability graph by using a non-maximum inhibition method; and setting a threshold value by a data set scale optimization method and carrying out binarization processing, thereby obtaining a final detection and segmentation result of the printing plate mesh point image to be detected.
The invention applies the deep learning technology to the printing plate dot image detection, and has the following advantages compared with the prior art:
1. different from the traditional printing plate dot image detection and segmentation method which relies on threshold setting, the method provided by the invention does not need to adjust parameters and can automatically run. After the deep learning network model is trained, no parameter is required to be set during reasoning, and model calculation result data post-processing can be automatically calculated through image characteristics.
2. Different from the traditional printing plate dot image detection and segmentation method which relies on artificial design features, the method provided by the invention can automatically extract the image features from the neural network, and no manual intervention is needed in the algorithm operation process. The deep learning network model has strong learning ability and high-efficiency feature expression ability, and can extract abstract semantic information from pixel-level original data layer by layer.
3. Compared with the traditional printing plate dot image detection and segmentation method, the method provided by the invention has the advantages that the segmentation effect is better, the boundary is more accurate, and various problems of the traditional method, such as continuity, robustness and the like, are solved. Experimental results show that the result evaluation indexes based on the deep learning method are obviously higher than those of the traditional dot image edge detection method.
4. Compared with the traditional printing plate dot image detection and segmentation method, the method provided by the invention can greatly improve the printing plate dot image detection and segmentation efficiency. The trained deep learning network model can run under the GPU, and the parallel computing efficiency of the deep learning network model is obviously higher than that of a traditional CPU computing method.
Drawings
FIG. 1 is a schematic flow chart of a printing plate dot image detection and segmentation method of the present invention.
FIG. 2 is a schematic diagram of a printing plate dot image and an artificially labeled edge image in the step of labeling image data according to the present invention.
FIG. 3 is a new dot-labeled image of a plate dot image labeled according to the present invention, generated using data enhancement techniques.
Fig. 4 is a diagram of a convolutional neural network architecture employed in the present invention.
Detailed Description
In order to better understand the technical solution of the present invention, the technical terms involved in the method and apparatus for detecting and segmenting the halftone dot image of the printing plate based on deep learning proposed by the present invention are explained as follows:
1. deep Learning (Deep Learning): deep learning is a new research direction in the field of machine learning, information is obtained by learning the internal rules and the representation levels of sample data, and the deep learning is greatly helpful for explaining data such as characters, images and sounds. The final aim of the method is to enable the machine to have the analysis and learning capability like a human, and to recognize data such as characters, images and sounds. Deep learning is a complex machine learning algorithm, and achieves the effect in speech and image recognition far exceeding the prior related art.
2. Dot (Printing Dot): the dots are the terms of printing industry, are basic units for expressing the gradation and color change of continuous tone images, are the basis for forming printing images and play a role in transmitting the tone of a layout. The state (size and shape) and behavior characteristics of the dots will affect whether the final print can correctly restore the original tone and color variations.
3. Edge Detection (Edge Detection): edge detection is a fundamental problem in image processing and computer vision, and the purpose of edge detection is to identify points in a digital image where brightness changes are significant. Significant changes in image attributes typically reflect significant events and changes in attributes.
4. Image Segmentation (Image Segmentation): image segmentation is a technique and process that divides an image into several specific regions with unique properties and proposes an object of interest. It is a key step from image processing to image analysis.
5. Convolutional Neural Network (CNN): the convolutional neural network is a feedforward neural network which comprises convolutional calculation and has a deep structure, and is one of representative algorithms of deep learning. The convolutional neural network has the characteristic learning ability and can carry out translation invariant classification on input information according to the hierarchical structure of the convolutional neural network.
6. Gradient Descent method (Gradient Descent): gradient descent is one type of iterative method that can be used to solve a least squares problem (both linear and non-linear). Gradient descent is one of the most commonly used methods when solving model parameters of machine learning algorithms, i.e. unconstrained optimization problems, and the other commonly used method is the least squares method.
7. Non-maximal inhibition method (Non-Maximum Suppression, NMS): non-maximum suppression is an edge thinning technique that can help suppress all gradient values except the local maximum (by setting them to 0), indicating the location with the strongest change in intensity value.
8. Dataset Scale optimization method (ODS): the globally optimal threshold setting method is to set the same threshold for all images in a data set, i.e. to select the threshold that maximizes the F-Score in the entire data set to be applied to all test pictures.
As shown in FIG. 1, the invention provides a printing plate dot image detection and segmentation method based on deep learning, which comprises the following steps:
the method comprises the following steps: the image data is annotated. That is, a certain number of edge image samples are drawn for the acquired high-resolution printing plate dot image.
As with previous deep learning algorithms, the present invention requires training the neural network using a data set to make the model tend to converge. Therefore, the acquired high-resolution printing plate dot image needs to be subjected to manual edge marking. Meanwhile, in order to improve the robustness of the algorithm model, a data enhancement method can be used for performing operations such as rotation, mirroring and shearing on the dot image, so that the scale of the data set is enlarged, and the condition of overfitting the model is effectively avoided.
Step two: and (5) constructing a network model. In order to detect and segment the printing plate dot images by using a deep learning model, a convolutional neural network is required to automatically learn rich hierarchical representation. Meanwhile, a loss function needs to be designed, the difference between a predicted value and an actual value is calculated, and the quality of model prediction is accurately measured.
Currently, a typical convolutional neural network infrastructure includes: leNet5, alexNet, VGGNet, *** InctionNet, resNet, and the like. The most used infrastructure for the edge detection algorithm is VGGNet. To improve detection efficiency, many lightweight neural networks are designed specifically for image edge detection. Meanwhile, the invention adopts a cross entropy loss function to calculate the error between the network output value and the true value, and considers that the label distribution imbalance is solved by adopting a weighting cross entropy loss method. The invention is not limited to a specific convolutional neural network structure, nor to a specific cross entropy loss function.
Step three: and training the network model. And training the network model constructed in the second step by using the image data marked in the first step until the loss function is converged, thereby obtaining an end-to-end deep learning edge detection model for the printing plate mesh point image.
At present, gradient descent is the most widely used optimization algorithm in a neural network, and in order to make up for the defects of a naive gradient descent algorithm, a plurality of neural network model training methods are proposed, including a random gradient descent method SGD, momentum gradient descent Momentum, optimizers RMSprop and Adam. The invention is not limited to a specific network model training method, and only needs to adopt a certain method to optimize the loss function of the network model and ensure the convergence of the network model.
Step four: a new image is detected. And (4) carrying out edge detection on the printing plate dot image to be detected by using the trained network model in the step three to obtain an edge probability graph of the printing plate dot image. Because the output of the network model is an edge probability graph, the network output result needs to be further corrected.
Step five: and correcting the detection result. And (4) refining the edge detection result by using a non-maximum suppression (NMS) method for the edge probability graph obtained in the step four. And setting a threshold value by an optimal data set scale (ODS) method, and carrying out binarization processing on the edge result after NMS refinement, thereby obtaining the final detection and segmentation result of the printing plate dot image to be detected.
Examples
The method comprises the following steps: the image data is annotated. In order to train the network model, 250 printing plate dot images are manually marked, 200 of the printing plate dot images are used for training the network model, and 50 of the printing plate dot images are used for evaluating the segmentation effect of model edge detection. Specifically, a printing plate to be collected is placed under a microscope, the focal length is adjusted to enable imaging to be clear, and a mesh point microscopic image of the printing plate is obtained by using an image collection system. Meanwhile, manually drawing the boundary of each dot in the printing plate image, and storing the result into a vector format. The sketched vectors can be converted into edge images after rasterization, and a certain number of training samples are manufactured according to the original images and the vectorized edge images. FIG. 2 shows a schematic representation of annotated image data, where 2a is a plate dot image and 2b is an artificially annotated edge image. The invention also adopts a data enhancement technology to carry out operations such as rotation, mirror image and cutting on the marked dot images, and the operations are added into the training set. Fig. 3 shows the enhancement result for the labeled image data of fig. 2, where 3a is the cut operation, 3b is the mirror operation, 3c is the rotate operation, and 3d is the brightness change operation.
Step two: and constructing a network model. The invention selects the VGG 16 model as the backbone network for feature extraction. To mimic the pyramidal structure, two or three stages are superimposed. The size (height x width) of the output feature map decreases with increasing number of stages, while the number of feature channels (number of filters) increases. The next layer is the aggregation of the dilation convolution. To capture the different levels of the receptive field, we used different dilation sizes and concatenated these deconvolution layers together for convolution operations to generate the edge probability map. Fig. 4 shows a network architecture employed by the present invention.
Step three: and training the network model. The invention adopts the pre-training weight of the HED algorithm to initialize the edge detection model of the printing plate dot image. Let the training set of the network model be
In which
To representNThe original input image is amplified and is taken>
Binary edge label representing an artificial mark of the input image, so->
，/>
The number of pixels representing one image.
Let all network parameter values of the VGG 16 model beWFor each boundary branch (side branches), the parameter value defining the boundary branch is
Then, the loss function of the network model is as shown in equation (1):
/>
wherein
The weight value of the loss function representing each boundary branch is set to @>
。
wherein the parameters
The weight value is used for balancing the unbalance of the positive and negative samples of the edge detection and is obtained by the ratio of the number of edge pixels and the number of non-edge pixels. />
，/>
，/>
Indicates the number of non-edge pixels, < > or >>
Indicating the number of edge pixels. />
Is shown asmA boundary is branched atjThe value of probability that the pixel prediction is an edge value is selectedSigmoidAnd calculating a function.
The optimizer that trains the network model uses a gradient descent method, as shown in equation (3).
Wherein:
is the loss function vs. the parameter->
Is based on the partial derivative of (4)>
Is the learning rate, and is the step size of each iteration update. The learning rate of the present invention starts from 0.001 and then is updated using a strategy in which the global training round number is gradually decreased, and each 10 epochs is multiplied by 0.1 until the loss function converges.
Step four: a new image is detected. The invention uses the trained edge detection model to carry out edge detection on the printing plate dot image to be segmented to obtain the edge probability graph of the printing plate dot image. When the image size of the printing plate lattice point shot by a microscope is larger, the image of the printing plate lattice point to be segmented can be divided into image blocks with smaller sizes to be respectively subjected to edge detection, and then detection results are spliced to form a complete edge probability graph. Meanwhile, the invention also adopts a multi-scale method to optimize the detection process. Specifically, the input printing plate dot image is scaled in three different scales [0.5,1.0,1.5] and input into the network; and finally, readjusting the outputs to the original sizes, and averaging the three output detection results to obtain a final edge probability graph.
Step five: and correcting the detection result. The present invention first refines the edge detection results using a non-maximum suppression (NMS) method. And then setting a threshold by adopting an optimal data set scale (ODS) method, namely selecting a fixed threshold for all printing plate dot images to enable the F-score value on the whole labeled image data set to be maximum. The F-score value is the harmonic mean of precision (P, precision) and recall (R, recall) as shown in equation (4).
Claims (5)
1. A printing plate dot image detection and segmentation method based on deep learning is characterized by comprising the following steps:
step one, labeling image data: drawing a certain number of edge image samples aiming at the acquired high-resolution printing plate dot images;
step two, constructing a network model: a network model is constructed by utilizing a convolutional neural network, and rich hierarchical representation is automatically learned; simultaneously designing a loss function, calculating the difference between a predicted value and an actual value, and using the difference to measure the result of model prediction;
step three, training a network model: training the network model constructed in the second step by using the image data marked in the first step until the loss function is converged to obtain an end-to-end deep learning edge detection model for the printing plate mesh point image;
step four, detecting a new image: performing edge detection on the printing plate dot image to be detected by using the trained network model in the third step to obtain an edge probability graph of the printing plate dot image;
step five, correcting the detection result: refining the edge detection result by using a non-maximum inhibition method for the edge probability graph obtained in the step four; and setting a threshold value by a data set scale optimization method and carrying out binarization processing, thereby obtaining a final detection and segmentation result of the printing plate mesh point image to be detected.
2. The method for detecting and segmenting the halftone dot image of the printing plate based on the deep learning of claim 1, wherein the image data is labeled in the first step by a specific process: carrying out artificial edge marking on the acquired high-resolution printing plate dot image; meanwhile, the data enhancement method is used for rotating, mirroring and shearing the dot images, and the data set scale is enlarged.
3. The deep learning-based printing plate dot image detecting and segmenting method according to claim 1, wherein the convolutional neural network infrastructure in the second step comprises: at least one of LeNet5, alexNet, VGGNet, *** Inception Net and ResNet.
4. The deep learning based printing plate dot image detecting and segmenting method according to claim 1, wherein the method for training the network model in the third step comprises at least one of a stochastic gradient descent method (SGD), a Momentum gradient descent Momentum, an optimizer (RMSprop) and Adam.
5. A printing plate dot image detection and segmentation device based on deep learning is characterized by comprising the following components:
the image data labeling module is used for drawing a certain number of edge image samples aiming at the acquired high-resolution printing plate dot images;
the network model building module is used for building a network model by utilizing a convolutional neural network and automatically learning rich hierarchical representation; simultaneously designing a loss function, and calculating the difference between a predicted value and an actual value so as to measure the result of model prediction;
the network model training module is used for training the constructed network model by using the marked image data until the loss function is converged to obtain an end-to-end deep learning edge detection model aiming at the printing plate dot image;
the image detection module is used for carrying out edge detection on the printing plate dot image to be detected by using the trained network model to obtain an edge probability graph of the printing plate dot image;
the detection result correction module is used for refining the edge detection result of the obtained edge probability graph by using a non-maximum inhibition method; and setting a threshold value by a data set scale optimization method and carrying out binarization processing, thereby obtaining a final detection and segmentation result of the printing plate mesh point image to be detected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310109402.4A CN115861297A (en) | 2023-02-14 | 2023-02-14 | Printing plate dot image detection and segmentation method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310109402.4A CN115861297A (en) | 2023-02-14 | 2023-02-14 | Printing plate dot image detection and segmentation method and device based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115861297A true CN115861297A (en) | 2023-03-28 |
Family
ID=85657975
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310109402.4A Pending CN115861297A (en) | 2023-02-14 | 2023-02-14 | Printing plate dot image detection and segmentation method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115861297A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116452827A (en) * | 2023-06-16 | 2023-07-18 | 青岛奥维特智能科技有限公司 | Method and system for detecting quality of ink printing surface based on computer vision |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109727232A (en) * | 2018-12-18 | 2019-05-07 | 上海出版印刷高等专科学校 | The detection method and equipment of the dot area percentage of printing plate |
| CN110570440A (en) * | 2019-07-19 | 2019-12-13 | 武汉珈和科技有限公司 | Image automatic segmentation method and device based on deep learning edge detection |
| CN110570427A (en) * | 2019-07-19 | 2019-12-13 | 武汉珈和科技有限公司 | Remote sensing image semantic segmentation method and device fusing edge detection |
| CN112258501A (en) * | 2020-11-10 | 2021-01-22 | 深圳拓匠印前科技有限公司 | Printing plate mesh point detection method, printing plate measuring instrument and readable storage medium |
| CN112308095A (en) * | 2019-07-30 | 2021-02-02 | 顺丰科技有限公司 | Picture preprocessing and model training method and device, server and storage medium |
| CN113255704A (en) * | 2021-07-13 | 2021-08-13 | 中国人民解放军国防科技大学 | Pixel difference convolution edge detection method based on local binary pattern |
-
2023
- 2023-02-14 CN CN202310109402.4A patent/CN115861297A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109727232A (en) * | 2018-12-18 | 2019-05-07 | 上海出版印刷高等专科学校 | The detection method and equipment of the dot area percentage of printing plate |
| CN110570440A (en) * | 2019-07-19 | 2019-12-13 | 武汉珈和科技有限公司 | Image automatic segmentation method and device based on deep learning edge detection |
| CN110570427A (en) * | 2019-07-19 | 2019-12-13 | 武汉珈和科技有限公司 | Remote sensing image semantic segmentation method and device fusing edge detection |
| CN112308095A (en) * | 2019-07-30 | 2021-02-02 | 顺丰科技有限公司 | Picture preprocessing and model training method and device, server and storage medium |
| CN112258501A (en) * | 2020-11-10 | 2021-01-22 | 深圳拓匠印前科技有限公司 | Printing plate mesh point detection method, printing plate measuring instrument and readable storage medium |
| CN113255704A (en) * | 2021-07-13 | 2021-08-13 | 中国人民解放军国防科技大学 | Pixel difference convolution edge detection method based on local binary pattern |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116452827A (en) * | 2023-06-16 | 2023-07-18 | 青岛奥维特智能科技有限公司 | Method and system for detecting quality of ink printing surface based on computer vision |
| CN116452827B (en) * | 2023-06-16 | 2023-08-15 | 青岛奥维特智能科技有限公司 | Method and system for detecting quality of ink printing surface based on computer vision |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110827251B (en) | Power transmission line locking pin defect detection method based on aerial image | |
| US10565479B1 (en) | Identifying and excluding blurred areas of images of stained tissue to improve cancer scoring | |
| CN108918536B (en) | Tire mold surface character defect detection method, device, equipment and storage medium | |
| CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
| US20200133182A1 (en) | Defect classification in an image or printed output | |
| US20200134382A1 (en) | Neural network training utilizing specialized loss functions | |
| CN111145163A (en) | Paper wrinkle defect detection method and device | |
| CN112307919B (en) | Improved YOLOv 3-based digital information area identification method in document image | |
| CN113888536B (en) | Printed matter double image detection method and system based on computer vision | |
| CN111127417B (en) | Printing defect detection method based on SIFT feature matching and SSD algorithm improvement | |
| CN110781901A (en) | Instrument ghost character recognition method based on BP neural network prediction threshold | |
| CN115496740A (en) | Lens defect detection method and system based on convolutional neural network | |
| CN112232351A (en) | License plate recognition system based on deep neural network | |
| CN115861297A (en) | Printing plate dot image detection and segmentation method and device based on deep learning | |
| CN114648806A (en) | Multi-mechanism self-adaptive fundus image segmentation method | |
| CN109685030A (en) | A kind of mug rim of a cup defects detection classification method based on convolutional neural networks | |
| CN116342536A (en) | Aluminum strip surface defect detection method, system and equipment based on lightweight model | |
| CN117422699A (en) | Highway detection method, highway detection device, computer equipment and storage medium | |
| CN112270404A (en) | Detection structure and method for bulge defect of fastener product based on ResNet64 network | |
| CN115063679B (en) | Pavement quality assessment method based on deep learning | |
| CN115359562A (en) | Sign language letter spelling recognition method based on convolutional neural network | |
| US20220366179A1 (en) | Assessment of image quality for optical character recognition using machine learning | |
| CN115457044A (en) | Pavement crack segmentation method based on class activation mapping | |
| CN114627081A (en) | Wheat imperfect grain identification method based on Mask R-CNN | |
| CN114092441A (en) | Product surface defect detection method and system based on dual neural network |
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 |