CN109087321B - Profile detection method based on visual path multi-level inhibition area synergistic effect - Google Patents

Profile detection method based on visual path multi-level inhibition area synergistic effect Download PDF

Info

Publication number
CN109087321B
CN109087321B CN201810575947.3A CN201810575947A CN109087321B CN 109087321 B CN109087321 B CN 109087321B CN 201810575947 A CN201810575947 A CN 201810575947A CN 109087321 B CN109087321 B CN 109087321B
Authority
CN
China
Prior art keywords
response
neuron
visual
knee
outer knee
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810575947.3A
Other languages
Chinese (zh)
Other versions
CN109087321A (en
Inventor
范影乐
蒋涯
张明琦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Dianzi University
Original Assignee
Hangzhou Dianzi University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hangzhou Dianzi University filed Critical Hangzhou Dianzi University
Priority to CN201810575947.3A priority Critical patent/CN109087321B/en
Publication of CN109087321A publication Critical patent/CN109087321A/en
Application granted granted Critical
Publication of CN109087321B publication Critical patent/CN109087321B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • General Physics & Mathematics (AREA)
  • Neurology (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Eye Examination Apparatus (AREA)
  • Image Processing (AREA)

Abstract

The invention relates to a contour detection method based on visual path multi-level inhibition area synergy. Firstly, constructing a direction-sensitive outer knee neuron array, recording the contour direction selected by an outer knee, and calculating the response of the outer knee neuron; the distance is taken as a factor to carry out quantization processing on the visual input of the classical receptive field and the inhibition zone of the external knee body neuron, the difference between the two is taken as the effective response of the inhibition zone, and the synergistic action parameters are obtained through dynamic rectification and power exponent normalization; the visual input of a primary visual cortex is obtained by weighting and fusing the responses of the external knee neurons through local windows, and the accurate contour response is detected; and calculating the response of the primary visual cortical neuron inhibition area to obtain the final contour response. The invention considers the step-by-step refined detection mechanism when the neurons in different levels sense the direction in the visual path, and simultaneously simulates the synergistic effect of the inhibition areas in different levels, thereby effectively improving the contour detection performance of the natural image.

Description

Profile detection method based on visual path multi-level inhibition area synergistic effect
Technical Field
The invention belongs to the field of machine vision, and mainly relates to a contour detection method based on visual path multi-level inhibition area synergistic effect.
Background
Contour detection is one of the early important links in image understanding or visual analysis, and the acquired contour features can effectively express the visual key details after redundancy removal. The difficulty of the contour detection task is mainly shown in two aspects of over-detection and under-detection, wherein the former is due to the interference of false contours such as texture noise, and the latter is due to the difference of the image on the contour contrast distribution. The traditional contour detection method is mainly based on the spatial jump of image information, so that a mathematical means based on operations such as difference or morphology is adopted, and good performance can be usually obtained under the condition of good image quality. However, the traditional method completely ignores the biological neural mechanism in visual perception, so that the requirement of detection performance is difficult to meet when facing complex tasks such as multi-level contour detection and the like. Although the current contour detection method based on the biological visual mechanism simulates the capability of biological visual to extract image contour features to a certain extent, when the method simulates the visual information flow processing process of a visual pathway, the method focuses more on the classical receptive field or inhibition zone of a certain level neuron on the visual pathway, and ignores the synergistic effect of the inhibition zones among different levels in the visual pathway, and the synergistic effect changes the isolation of each level of a visual perception model, and can fully play the important role played by each level in perception by the characteristics of overall synergy.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a contour detection method based on the synergistic effect of a visual path multi-level inhibition area.
This patent considers: (1) the primary visual cortex neuron has a direction selection characteristic on visual excitation, a plurality of discretization angles are usually set in the traditional detection method, then the angle with a response extreme value is determined as a contour direction, the detection performance is closely related to the degree of angle discretization, the dense discretization angle obviously consumes the limited resources of the detection system, and the dispersed discretization angle seriously influences the detection accuracy. (2) When the traditional method is used for simulating the direction selection characteristic of the primary visual cortical neuron, the comprehensive consideration of angles and scales is generally only carried out in the hierarchy, the synergistic effect of a preceding link on the primary visual cortical neuron is not considered, and the relevance between the hierarchies can help to transition from local details to the description of the whole structure. (3) Traditional methods typically use gaussian difference models to characterize the zone of inhibition within a single level of the primary visual cortex, while ignoring the neurovisual mechanisms of multi-level zone of inhibition synergy on the visual pathway. The cooperation enables the transmission and processing of the visual information flow on the visual path to form an effective whole, and facilitates the expansion and fusion of the contour detection from the local and global visual angles.
Therefore, the invention provides a contour detection method based on the synergistic effect of the visual pathway multi-level inhibition areas, starting from a step-by-step fine detection mechanism when the neurons at different levels in the visual pathway sense the direction, and simulating the synergistic effect of the different layers of the inter-level inhibition areas, so as to fully play the important role of the visual characteristics in contour detection.
The method mainly comprises the following steps:
firstly, constructing an outer knee neuron array corresponding to pixel points one by one; and then simulating a direction selection mechanism of the outer knee body neurons, setting k discrete directions to be detected, and calculating the response intensity of each direction and the ratio of the direction response intensity to the sum of all the direction response intensities aiming at the outer knee body neurons corresponding to each pixel point I (x, y). If the maximum value of the ratio exceeds the threshold value, recording the direction corresponding to the maximum value, and regarding the direction as the optimal response direction ID1(x, y); if all the occupation ratios do not reach the threshold value, the directions corresponding to the first two numerical values of the occupation ratios, namely the optimal response direction and the suboptimal response direction, are recorded at the same time and are respectively recorded as ID1(x, y) and ID2(x, y). Finally, the maximum value of the direction response is used as the response of the external knee body neuron.
Step (2) calculating the synergistic action parameter w of the outer knee body-primary visual cortex inhibition areaLGN(x,y)。
Firstly, obtaining the visual input of the classical receptive field of the outer knee body neuron by utilizing a Gaussian function, obtaining the visual input of an inhibition area of the outer knee body neuron by utilizing a double-Gaussian difference function, and respectively carrying out quantization processing on the two visual inputs based on distance factors; then the difference of the two visual inputs after the quantization processing is used as the effective response of the outer knee neuron inhibition area, and the effective response is subjected to dynamic half-wave rectification. Finally, performing power exponent normalization on the effective response after the dynamic half-wave rectification; setting the power exponent normalization result as a synergistic action parameter w for regulating the response of the primary visual cortex neuron inhibition regionLGN(x,y)。
And (3) constructing a primary visual cortex neuron array with the same size as the outer knee body neuron array, and determining the visual input U (x, y) of each neuron of the primary visual cortex.
Firstly, a local window is constructed to serve as a receptive field of a certain neuron of a primary visual cortex, then synaptic connection weights of all outer knee somatic neurons in the receptive field and the neuron of the primary visual cortex are calculated, and finally responses of all outer knee somatic neurons in the window are fused to obtain visual input U (x, y) of the neuron of the primary visual cortex.
And (4) realizing a direction refined detection mechanism of the primary visual cortex neuron.
If the outer knee only records the ID in the step (1)1(x, y), respectively setting q discrete directions in bilateral symmetry in the optimal response direction based on the fixed discrete angle interval; if the outer knee simultaneously records ID1(x, y) and ID2(x, y), then p discrete directions are set in total between the optimal and suboptimal response directions. The directional responses of the primary visual cortical neurons at the visual input U (x, y) are detected for the 2q or p discrete directions, respectively. Taking the maximum value of the directional response as the accurate contour response E of the primary visual cortex detectionV1(x,y)。
And (5) firstly calculating the response Inh (x, y) of the primary visual cortex inhibition zone fused with the spatial scale factor. Then utilizing the synergistic action parameter w obtained in the step (2)LGNAdjusting Inh (x, y) by (x, y), and finally using the adjustment result to suppress the accurate contour response EV1(x, y) and taking the suppressed response as a final contour response E (x, y).
The invention has the following beneficial effects:
1. the detection mechanism that becomes more meticulous step by step when simulation multilevel neuron detects the profile direction has reduced the coupling nature of detecting profile direction and angle discretization degree, has satisfied the continuity of direction to a certain extent simultaneously, and the finite resource of make full use of detecting system detects the profile direction more fast more accurately.
2. The method simulates a synergistic action mechanism of a multi-level inhibition zone, constructs an outer knee body-primary visual cortex inhibition zone synergistic action model, extracts synergistic action parameters based on effective response of the outer knee body inhibition zone, adjusts the inhibition zone strength of the primary visual cortex, weakens self-inhibition between contours, protects a real contour while removing non-contour textures and false edges, and better accords with an information processing mode of the inhibition zone synergistic action in a visual pathway.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
With reference to the attached drawing 1, the specific implementation steps of the invention are as follows:
firstly, constructing an external knee neuron array corresponding to pixel points one by one. Then, a direction selection mechanism of the external knee neurons is simulated, and k (default is 6) discrete directions to be detected as shown in the formula (1) are set. Then, for the outer knee body neuron corresponding to each pixel point I (x, y), a specific direction theta is obtained as shown in formula (2)iIntensity of response eLGN(x,y；θi,σl). The ratio d (x, y; theta) of the response intensity of the direction and the sum of the response intensities of all the directions is calculated simultaneouslyi) As shown in formula (3).
Figure RE-GDA0001830908270000041
Figure RE-GDA0001830908270000042
Figure RE-GDA0001830908270000043
Where is the convolution symbol, I (x, y) is the input image,
Figure RE-GDA0001830908270000044
Figure RE-GDA0001830908270000045
σlis the size of the classical receptive field of the outer knee neurons, and is set to 2 by default. Maximum value d of ratiomax(x,y；θi) If the threshold value is exceeded (default is 0.5), the direction corresponding to the maximum value is recorded and is regarded as the optimal response direction ID1(x, y); if all the occupation ratios do not reach the threshold value, recording the occupation ratios beforeThe directions corresponding to the two values, i.e. the optimal response direction and the suboptimal response direction, are respectively marked as ID1(x, y) and ID2(x, y). Finally, the maximum value of the direction response is taken as the outer knee body neuron response E corresponding to the pixel point I (x, y)LGN(x, y) is represented by the formula (4).
ELGN(x,y)＝max{eLGN(x,y；θi,σl)|i＝0,1,...k-1} (4)
Step (2) calculating the synergistic action parameter w of the outer knee body-primary visual cortex inhibition areaLGN(x, y). Firstly, the visual input of the classical receptive field of the outer knee somatic neuron is obtained by utilizing a Gaussian function, the visual input of the inhibition area of the outer knee somatic neuron is obtained by utilizing a double Gaussian difference function, and the two visual inputs are respectively subjected to quantization processing based on distance factors.
Figure RE-GDA0001830908270000051
And
Figure RE-GDA0001830908270000052
the visual input is respectively the visual input after the quantization treatment of the classical receptive field and the inhibition zone, and the definitions are shown in formulas (5) to (8).
Figure RE-GDA0001830908270000053
Figure RE-GDA0001830908270000054
Figure RE-GDA0001830908270000055
DoG+(m,n；σl,ρl)＝max{gaus(m,n,ρlσl)-gaus(m,n,σl),0} (8)
Where ρ islThe ratio of the size of the outer knee body neuron inhibition area to the size of the classical receptive field is set as 4 by default, m and n represent the interior of the classical receptive fieldThe offset between visual input I (x + m, y + n) and I (x, y). Then will be
Figure RE-GDA0001830908270000056
Difference of (2)
Figure RE-GDA0001830908270000057
As an effective response of the external knee neuron inhibition area, and dynamically half-wave rectifying delta phi (x, y), namely taking the mean value mean of the delta phi (x, y) as a threshold value, and performing dynamic half-wave rectification on the delta phi (x, y) to obtain a voltage value lower than the threshold value
Figure RE-GDA0001830908270000058
And setting 0, keeping the value not lower than the threshold unchanged as shown in a formula (9).
Figure RE-GDA0001830908270000059
Finally, carrying out power exponent normalization on the delta phi (x, y) after dynamic half-wave rectification, and adjusting the synergistic action parameter w of the primary visual cortex neuron inhibition area responseLGN(x, y) is set to this normalized result as shown in equation (10).
Figure RE-GDA0001830908270000061
Wherein, delta is an effective response adjusting parameter of the inhibition zone, and the default value is 2.
And (3) constructing a primary visual cortex neuron array with the same size as the outer knee body neuron array, and determining the visual input U (x, y) of each neuron of the primary visual cortex. Firstly, a local window W is constructedxyAs the receptive field of a certain neuron in the primary visual cortex, calculating the synaptic connection weight w (x) of all external knee somatic neurons in the receptive field and the neuron in the primary visual cortext,yt) As shown in formula (11):
Figure RE-GDA0001830908270000062
where r is the local window radius (default to 3), (x)t,yt) Expressed as the t-th outer knee neuron centered at (x, y) in the window, and finally, the responses of all outer knee neurons in the window are fused to obtain the visual input U (x, y) of the neuron in the primary visual cortex, as shown in formula (12).
Figure RE-GDA0001830908270000063
Where μ is the sum of all outer knee neuron responses within the window.
And (4) realizing a direction refined detection mechanism of the primary visual cortex neuron. If the outer knee only records the ID in the step (1)1(x, y), respectively setting q discrete directions in bilateral symmetry in the optimal response direction based on the fixed discrete angle interval s, as shown in formula (13); if the outer knee simultaneously records ID1(x, y) and ID2(x, y), then a total of p discrete directions are set equidistant between the optimal and suboptimal response directions, as shown in equation (14). Default s is 5 °, p is 3, and q is 6.
θh＝ID1(x,y)+(h-q)×s,h＝0,1,...2q-1 (13)
Figure RE-GDA0001830908270000071
Aiming at the 2q or p discrete directions, detecting the visual input U (x, y) of the primary visual cortex neuron in the direction thetahDirectional response of time eV1(x,y；θh,σv) As shown in formula (15).
Figure RE-GDA0001830908270000072
Wherein
Figure RE-GDA0001830908270000073
σvIs the primary optic cortex neuron channelThe size of the classical reception field is set to 4 by default. Using the maximum value of the directional response as the accurate contour response E detected by the primary visual cortexV1(x, y). As shown in equation (16).
EV1(x,y)＝max{eV1(x,y；θh,σv)|h＝0,1,...2q-1OR h＝0,1,...p-1} (16)
And (5) firstly, calculating the response Inh (x, y) of the primary visual cortical neuron inhibition region fused with the spatial scale factors, as shown in formulas (17) and (18).
Figure RE-GDA0001830908270000074
Figure RE-GDA0001830908270000075
Wherein, | DoG+(x,y；σv,ρv) L is DoG+(x,y；σv,ρv) L1 norm, ρvThe ratio of the primary visual cortex neuron inhibition area to the classical receptive field is 4 as a default value, and then the synergistic action parameter w obtained in the step (2) is utilizedLGN(x, y) regulates Inh (x, y). Finally, the adjusting result is used for restraining the accurate contour response EV1(x, y) and taking the suppressed response as a final contour response E (x, y). As shown in equation (19).
E(x,y)＝EV1(x,y)-α·wLGN(x,y)·Inh(x,y) (19)
Wherein alpha is a parameter for adjusting the overall inhibition intensity of the inhibition zone and is 1.0 as default.

Claims (1)

1. A contour detection method based on visual path multi-level inhibition area synergy is characterized by comprising the following steps:
firstly, constructing an outer knee neuron array corresponding to pixel points one by one; then simulating the direction selection mechanism of external knee neurons, and setting the direction to be detectedkA discrete direction for each pixel pointI (x, y) corresponding to the external knee neurons, and calculating the response intensity of each direction and the ratio of the response intensity of the direction to the sum of the response intensities of all directions; if the maximum value of the ratio exceeds the threshold value, recording the direction corresponding to the maximum value, and regarding the direction as the optimal response direction ID1(x, y); if all the occupation ratios do not reach the threshold value, the directions corresponding to the first two numerical values of the occupation ratios, namely the optimal response direction and the suboptimal response direction, are recorded at the same time and are respectively recorded as ID1(x, y) and ID2(x, y); finally, the maximum value of the direction response is used as the response of the external knee body neuron;
step (2) calculating the synergistic action parameter w of the outer knee body-primary visual cortex inhibition areaLGN(x,y)；
Firstly, obtaining the visual input of the classical receptive field of the outer knee somatic neuron by using a Gaussian function, obtaining the visual input of an inhibition area of the outer knee somatic neuron by using a double Gaussian difference function, and respectively carrying out quantitative processing on the two visual inputs based on distance factors; then taking the difference of the two visual inputs after the quantization processing as the effective response of the outer knee body neuron inhibition area, and carrying out dynamic half-wave rectification on the effective response; finally, performing power exponent normalization on the effective response after the dynamic half-wave rectification; setting the power exponent normalization result as a synergistic action parameter w for regulating the response of the primary visual cortex neuron inhibition regionLGN(x,y)；
Constructing a primary visual cortex neuron array with the same size as the outer knee body neuron array, and determining the visual input U (x, y) of each neuron of the primary visual cortex;
firstly, constructing a local window as a receptive field of a certain neuron of a primary visual cortex, then calculating synaptic connection weights of all outer knee somatic neurons in the receptive field and the neuron of the primary visual cortex, and finally fusing the responses of all outer knee somatic neurons in the window to obtain visual input U (x, y) of the neuron of the primary visual cortex;
step (4), realizing a direction refined detection mechanism of primary visual cortical neurons;
if the outer knee only records the ID in the step (1)1(x, y), then left of the optimal response direction based on a fixed discrete angle intervalEach setting of right symmetryqA discrete direction; if the outer knee simultaneously records ID1(x, y) and ID2(x, y), then co-setting between optimal and suboptimal response directionspA discrete direction; to 2qOr ispDiscrete directions, which respectively detect the directional response of primary visual cortical neurons when in visual input U (x, y); taking the maximum value of the directional response as the accurate contour response E of the primary visual cortex detectionV1(x,y)；
Step (5), firstly, calculating the response Inh (x, y) of a primary visual cortex inhibition zone fused with spatial scale factors; then utilizing the synergistic action parameter w obtained in the step (2)LGNAdjusting Inh (x, y) by (x, y), and finally using the adjustment result to suppress the accurate contour response EV1(x, y) and taking the suppressed response as a final contour response E (x, y).
CN201810575947.3A 2018-06-06 2018-06-06 Profile detection method based on visual path multi-level inhibition area synergistic effect Active CN109087321B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810575947.3A CN109087321B (en) 2018-06-06 2018-06-06 Profile detection method based on visual path multi-level inhibition area synergistic effect

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810575947.3A CN109087321B (en) 2018-06-06 2018-06-06 Profile detection method based on visual path multi-level inhibition area synergistic effect

Publications (2)

Publication Number Publication Date
CN109087321A CN109087321A (en) 2018-12-25
CN109087321B true CN109087321B (en) 2021-08-10

Family

ID=64839425

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810575947.3A Active CN109087321B (en) 2018-06-06 2018-06-06 Profile detection method based on visual path multi-level inhibition area synergistic effect

Country Status (1)

Country Link
CN (1) CN109087321B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109978898B (en) * 2019-02-01 2023-07-18 广西科技大学 Contour detection method based on vector field energy calculation
CN111402285B (en) * 2020-01-16 2023-08-08 杭州电子科技大学 Contour detection method based on visual mechanism dark edge enhancement

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102201120A (en) * 2011-06-17 2011-09-28 电子科技大学 Multifeature-based target object contour detection method
CN103473759A (en) * 2013-06-24 2013-12-25 南京理工大学 Low-light-level image significant contour extraction method of WKPCA homogeneity degree correction nCRF inhibition
CN104778696A (en) * 2015-04-13 2015-07-15 杭州电子科技大学 Image edge grading-detection method based on visual pathway orientation sensitivity
CN106127740A (en) * 2016-06-16 2016-11-16 杭州电子科技大学 A kind of profile testing method based on the association of visual pathway many orientation of sensory field

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102201120A (en) * 2011-06-17 2011-09-28 电子科技大学 Multifeature-based target object contour detection method
CN103473759A (en) * 2013-06-24 2013-12-25 南京理工大学 Low-light-level image significant contour extraction method of WKPCA homogeneity degree correction nCRF inhibition
CN104778696A (en) * 2015-04-13 2015-07-15 杭州电子科技大学 Image edge grading-detection method based on visual pathway orientation sensitivity
CN106127740A (en) * 2016-06-16 2016-11-16 杭州电子科技大学 A kind of profile testing method based on the association of visual pathway many orientation of sensory field

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Multifeature-Based Surround Inhibition Improves Contour Detection in Natural Images;Kai-Fu Yang等;《IEEE Transactions on Image Processing》;20141001;第1-13页 *
基于多视觉特征的非经典感受野模型及应用研究;杨开富;《中国优秀硕士学位论文全文数据库》;20130115;I138-1296 *
基于突触连接视通路方位敏感的图像分级边缘检测方法;王典等;《中国生物医学工程学报》;20151031;第522-532页 *

Also Published As

Publication number Publication date
CN109087321A (en) 2018-12-25

Similar Documents

Publication Publication Date Title
CN108095716B (en) Electrocardiosignal detection method based on confidence rule base and deep neural network
Melin et al. Edge-detection method for image processing based on generalized type-2 fuzzy logic
CN107481188A (en) A kind of image super-resolution reconstructing method
CN105741267B (en) The multi-source image change detecting method of cluster guidance deep neural network classification
CN106951825A (en) A kind of quality of human face image assessment system and implementation method
CN111652822B (en) Single image shadow removing method and system based on generation countermeasure network
CN106127740B (en) One kind being based on the associated profile testing method of the more orientation of sensory field of visual pathway
CN111145145B (en) Image surface defect detection method based on MobileNet
CN112990677B (en) Teaching system, computer equipment and storage medium based on artificial intelligence
WO2008016109A1 (en) Learning data set optimization method for signal identification device and signal identification device capable of optimizing the learning data set
Chakraborty et al. Contrast optimization using elitist metaheuristic optimization and gradient approximation for biomedical image enhancement
CN109087321B (en) Profile detection method based on visual path multi-level inhibition area synergistic effect
CN111259735B (en) Single-person attitude estimation method based on multi-stage prediction feature enhanced convolutional neural network
CN112819063B (en) Image identification method based on improved Focal loss function
CN107633527A (en) Target tracking method and device based on full convolutional neural networks
CN111798463B (en) Method for automatically segmenting multiple organs in head and neck CT image
CN110443790B (en) Cartilage identification method and system in medical image
CN111104339B (en) Software interface element detection method, system, computer equipment and storage medium based on multi-granularity learning
CN107024480A (en) A kind of stereoscopic image acquisition device
CN115861625A (en) Self-label modifying method for processing noise label
Liu et al. A novel image segmentation algorithm based on visual saliency detection and integrated feature extraction
CN113192092B (en) Contour detection method for simulating fusion of properties of receptor field of XYW cell
Sehgal Edge detection techniques in digital image processing using Fuzzy Logic
KR20230077367A (en) Artificial intelligence apparatus for determining deterioration level based on image of transmission tower and method thereof
CN111968140A (en) Contour detection method based on classical receptive field fixation micromotion mechanism

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
GR01 Patent grant
GR01 Patent grant