CN105512664A - Image recognition method and device - Google Patents
Image recognition method and device Download PDFInfo
- Publication number
- CN105512664A CN105512664A CN201510881619.2A CN201510881619A CN105512664A CN 105512664 A CN105512664 A CN 105512664A CN 201510881619 A CN201510881619 A CN 201510881619A CN 105512664 A CN105512664 A CN 105512664A
- Authority
- CN
- China
- Prior art keywords
- image
- identified
- histograms
- oriented gradients
- identified region
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an image recognition method and device. The method comprises the steps of extracting the histogram characteristics of the oriented gradient of a to-be-recognized image; and recognizing the histogram characteristics of the oriented gradient of the to-be-recognized image by means of an adaboost classifier to obtain a recognition result. The adaboost classifier is pre-obtained through the classification-based training step of the histogram characteristics of the oriented gradient of a sample image based on the adaboost training process. The histogram characteristics of the oriented gradient of the image are adopted as characteristics that are to be recognized during the adaboost training process. In this way, not only the fast property of the adaboost is utilized, but also the good contour characteristics description property of the histogram of oriented gradients is utilized at the same time. Therefore, the fast and high-precision image recognition is realized through extracting the histogram characteristics of the oriented gradient of the to-be-recognized image and recognizing the histogram characteristics of the oriented gradient of the to-be-recognized image by means of the adaboost classifier.
Description
Technical field
The disclosure relates to image processing field, particularly relates to image-recognizing method and device.
Background technology
Image recognition, refers to and utilizes computer technology to process image, analyze and understand, to identify the technology of various object.Such as, in some correlation techniques, can the human body in image be identified.But these correlation techniques or speed are slow or accuracy of identification is not high.The needs of people to image recognition rate and precision can not be met.
Summary of the invention
For overcoming Problems existing in correlation technique, the disclosure provides a kind of image-recognizing method and device.
According to the first aspect of disclosure embodiment, provide a kind of image-recognizing method.The method comprises: the histograms of oriented gradients feature extracting image to be identified; Call the histograms of oriented gradients feature of adaboost sorter to described image to be identified to identify, obtain recognition result, wherein, described adaboost sorter trains the histograms of oriented gradients feature of flow process to sample image to carry out classification based training acquisition based on adaboost in advance.
The technical scheme that disclosure embodiment provides can comprise following beneficial effect: the technical scheme that disclosure embodiment provides trains the histograms of oriented gradients feature of flow process to sample image to carry out classification based training based on adaboost in advance, obtain detection speed adaboost sorter faster, due to using histograms of oriented gradients feature as adaboost training in feature, both make use of adaboost feature fast, merge again the feature of the reasonable description contour feature of histograms of oriented gradients, therefore the histograms of oriented gradients feature of image to be identified is extracted, call the histograms of oriented gradients feature that adaboost sorter treats recognition image to identify, achieve fast, high-precision image recognition.
According to a kind of possible embodiment of disclosure embodiment first aspect, described sample image comprises: human sample image and non-human sample image.Described adaboost sorter is the adaboost sorter for human detection.Described recognition result is the human bioequivalence result of described image to be identified.
The technical scheme that disclosure embodiment provides can comprise following beneficial effect: because disclosure embodiment trains flow process to carry out classification based training based on adaboost, obtain detection speed adaboost sorter faster, and, based on histograms of oriented gradients feature, feature interpretation is carried out to human sample image and non-human sample image, both make use of adaboost feature fast, merge again the feature of the reasonable description contour feature of histograms of oriented gradients, call the histograms of oriented gradients feature that adaboost sorter treats recognition image to identify, can meet human bioequivalence speed fast, the needs that accuracy of identification is high.
According to a kind of possible embodiment of disclosure embodiment first aspect, the histograms of oriented gradients of described sample image is characterized as the histograms of oriented gradients feature mapped through linear discriminant analysis.Described method also comprises: carry out linear discriminant analysis mapping to the histograms of oriented gradients feature of described image to be identified, obtains the histograms of oriented gradients feature that described image to be identified maps through linear discriminant analysis.The histograms of oriented gradients feature of the described adaboost of calling sorter to described image to be identified identifies, obtain recognition result to comprise: call described adaboost sorter and identify the histograms of oriented gradients feature that described image to be identified maps through linear discriminant analysis, obtain recognition result.
The technical scheme that disclosure embodiment provides can comprise following beneficial effect: the present embodiment is by carrying out linear discriminant analysis mapping to histograms of oriented gradients feature, can make the histograms of oriented gradients feature of renewal that obtains after mapping can the contour feature of Description Image more exactly, thus improve accuracy of identification.
According to a kind of possible embodiment of disclosure embodiment first aspect, described method also comprises: obtain original image; By multiple scaling, convergent-divergent is carried out to described original image, obtains multiple zoomed images; Multiple zoomed images described are carried out to the division of identified region, obtain described multiple zoomed images each identified region separately, each identified region is an image to be identified.Wherein, the histograms of oriented gradients feature of the described adaboost of calling sorter to described image to be identified identifies, obtain recognition result to comprise: call the histograms of oriented gradients feature of described adaboost sorter to each identified region and identify, obtain the recognition result of each identified region.Described method also comprises: according to the scaling of each identified region place zoomed image relative to original image, each identified region is adjusted to the size relative to original image with same zoom ratio, by each identified region after adjustment is compared, find out between different identified region, meet the identified region of coincidence condition, the recognition result of the identified region of satisfied coincidence condition is merged, obtains the final recognition result of each identified region.
The technical scheme that disclosure embodiment provides can comprise following beneficial effect: in the present embodiment, have employed and by multiple scaling, convergent-divergent is carried out to original image, Region dividing is carried out to the image after convergent-divergent, the region marked off is identified, a kind of like this recognition method of pyramid level traverse scanning, make the different identified regions of original image can the mating with sample image of maximum probability, thus can the image that actual scene comprises multiple identification object be identified, improve accuracy of identification.
According to a kind of possible embodiment of disclosure embodiment first aspect, described acquisition original image comprises: the shooting image obtaining actual scene from picture pick-up device; The model of place of the described actual scene that utilization is set up in advance removes the background in described shooting image, obtains described original image.
The technical scheme that disclosure embodiment provides can comprise following beneficial effect: because this embodiment can remove the background of shooting image to obtain original image, therefore, the technical scheme that disclosure embodiment provides can identify the image that actual scene comprises background, improves accuracy of identification.
According to a kind of possible embodiment of disclosure embodiment first aspect, when described recognition result is the human bioequivalence result of described image to be identified, described method also comprises: add up the final recognition result of each identified region described, obtains the people flow rate statistical result in described actual scene.
The technical scheme that disclosure embodiment provides can comprise following beneficial effect: in the present embodiment, owing to carrying out convergent-divergent to original image by multiple scaling, Region dividing is carried out to the image after convergent-divergent, to the histograms of oriented gradients feature in the region marked off, the adaboost sorter being used for human detection is utilized to identify, therefore, it is possible to carry out human bioequivalence to the image that actual scene comprises multiple human body fast, accurately, obtain the people flow rate statistical result in actual scene.
According to the second aspect of disclosure embodiment, provide a kind of pattern recognition device.This device comprises: characteristic extracting module, is configured to the histograms of oriented gradients feature extracting image to be identified; Identification module, be configured to call the histograms of oriented gradients feature of adaboost sorter to the image described to be identified that described characteristic extracting module is extracted identify, obtain recognition result, wherein, described adaboost sorter trains the histograms of oriented gradients feature of flow process to sample image to carry out classification based training acquisition based on adaboost in advance.
According to a kind of possible embodiment of disclosure embodiment second aspect, described sample image comprises: human sample image and non-human sample image.Described adaboost sorter is the adaboost sorter for human detection.Described recognition result is the human bioequivalence result of described image to be identified.
According to a kind of possible embodiment of disclosure embodiment second aspect, the histograms of oriented gradients of described sample image is characterized as the histograms of oriented gradients feature mapped through linear discriminant analysis.Described characteristic extracting module is also configured to carry out linear discriminant analysis mapping to the histograms of oriented gradients feature of described image to be identified, obtains the histograms of oriented gradients feature that described image to be identified maps through linear discriminant analysis.Described identification module is configured to call adaboost sorter and identifies the histograms of oriented gradients feature through linear discriminant analysis mapping that described characteristic extracting module is extracted, and obtains recognition result.
According to a kind of possible embodiment of disclosure embodiment second aspect, described device also comprises: original image acquisition module, is configured to obtain original image.Zoom module, the original image be configured to described original image acquisition module obtains carries out convergent-divergent by multiple scaling, obtains multiple zoomed images.Region dividing module, is configured to the division multiple zoomed images that described Zoom module obtains being carried out to identified region, and obtain described multiple zoomed images each identified region separately, each identified region is an image to be identified.Wherein, described identification module is configured to call the histograms of oriented gradients feature of described adaboost sorter to each identified region that described Region dividing module obtains and identifies, obtains the recognition result of each identified region.Described device also comprises: size adjustment module, to be configured to according to each identified region place zoomed image, relative to the scaling of original image, each identified region be adjusted to the size relative to original image with same zoom ratio.Comparison module, is configured to, by being compared by each identified region after described size adjustment module adjustment, find out between different identified region, meet the identified region of coincidence condition.Merge module, be configured to the recognition result of the identified region of satisfied coincidence condition to merge, obtain the final recognition result of each identified region.
According to a kind of possible embodiment of disclosure embodiment second aspect, described original image acquisition module comprises: shooting Image Acquisition submodule, is configured to the shooting image obtaining actual scene from picture pick-up device.Background detection submodule, is configured to utilize the model of place of the described actual scene set up in advance to remove background in described shooting image, obtains described original image.
According to a kind of possible embodiment of disclosure embodiment second aspect, described device also comprises: statistical module, be configured to when described recognition result is the human bioequivalence result of described image to be identified, the final recognition result of each identified region described is added up, obtains the people flow rate statistical result in described actual scene.
According to the third aspect of disclosure embodiment, provide a kind of pattern recognition device.This device comprises: processor; For the storer of storage of processor executable instruction; Wherein, described processor is configured to: the histograms of oriented gradients feature extracting image to be identified; Call the histograms of oriented gradients feature of adaboost sorter to described image to be identified to identify, obtain recognition result, wherein, described adaboost sorter trains the histograms of oriented gradients feature of flow process to sample image to carry out classification based training acquisition based on adaboost in advance.
Should be understood that, it is only exemplary and explanatory that above general description and details hereinafter describe, and can not limit the disclosure.
Accompanying drawing explanation
Accompanying drawing to be herein merged in instructions and to form the part of this instructions, shows and meets embodiment of the present disclosure, and is used from instructions one and explains principle of the present disclosure.
Fig. 1 is the structural representation of a kind of implementation environment according to an exemplary embodiment.
Fig. 2 is the process flow diagram of the image-recognizing method according to an exemplary embodiment.
Fig. 3 is the process flow diagram of the image-recognizing method according to another exemplary embodiment.
Fig. 4 is the process flow diagram of the image-recognizing method according to another exemplary embodiment.
Fig. 5 is the process flow diagram of the image-recognizing method Gen Ju an exemplary embodiment again.
Fig. 6 is the original image schematic diagram according to an exemplary embodiment.
Fig. 7 is the process flow diagram of the image-recognizing method according to another exemplary embodiment.
Fig. 8 is the block diagram of the pattern recognition device according to an exemplary embodiment.
Fig. 9 is the block diagram of the pattern recognition device according to another exemplary embodiment.
Figure 10 is the block diagram of a kind of pattern recognition device according to an exemplary embodiment.
Embodiment
Here will be described exemplary embodiment in detail, its sample table shows in the accompanying drawings.When description below relates to accompanying drawing, unless otherwise indicated, the same numbers in different accompanying drawing represents same or analogous key element.Embodiment described in following exemplary embodiment does not represent all embodiments consistent with the disclosure.On the contrary, they only with as in appended claims describe in detail, the example of apparatus and method that aspects more of the present disclosure are consistent.
Fig. 1 is the structural representation of a kind of implementation environment according to an exemplary embodiment.As shown in Figure 1, this implementation environment can comprise: picture pick-up device 110 and server 120.
Such as, picture pick-up device 110 can be intelligent camera or the intelligent video camera head with memory function and processing capacity.Server 120 may be used for the data storing, process and/or send picture pick-up device 110.Picture pick-up device 110 can be connected with server 120, be intercomed mutually by network.This network such as can include but not limited to: the network such as WiFi (WirelessFidelity, Wireless Fidelity), 2G, 3G, 4G.
Fig. 2 is the process flow diagram of the image-recognizing method according to an exemplary embodiment.The method can be applied to picture pick-up device 110 shown in Fig. 1 or server 120.As shown in Figure 2, the method can comprise:
In step 210, the histograms of oriented gradients feature of image to be identified is extracted.
Wherein, histograms of oriented gradients (HistogramofOrientedGradient, hog) feature is a kind of feature interpretation operator being used for carrying out object detection in computer vision and image procossing.It carrys out constitutive characteristic by the gradient orientation histogram of calculating and statistical picture regional area.
Such as, the process extracting the histograms of oriented gradients feature of image to be identified comprises: treat recognition image and be normalized, calculate the histogram of gradients of normalized sample image, each cell block is carried out to the projection of default weight to histogram of gradients, contrast normalization is carried out for the cell block in each overlapping block, histogram vectors in all overlapping blocks is combined as histograms of oriented gradients proper vector, is histograms of oriented gradients feature.
In a step 220, call the histograms of oriented gradients feature of adaboost sorter to described image to be identified to identify, obtain recognition result, wherein, described adaboost sorter trains the histograms of oriented gradients feature of flow process to sample image to carry out classification based training acquisition based on adaboost in advance.
Wherein, adaboost (AdaptiveBoosting, self-adaptation strengthens) is a kind of machine learning method.The training flow process of this machine learning method is a kind of iterative algorithm, takes turns in iteration add a new Weak Classifier, until reach certain predetermined enough little error rate at each.Training flow process comprises: each training sample is endowed a weight, show that it is selected into the probability of training set by certain sorter, if certain sample point is classified exactly, so under construction in a training set, its selected probability is just lowered, on the contrary, if certain sample point is not classified exactly, so its weight is just improved.By such training flow process, adaboost can focus on more indistinguishable sample.Such as, the weight of each sample can be made at first equal, for kth time iterative operation, choose sample point according to weight, and then training classifier C
k.Then according to sorter C
k, improve by the weight of the sample of its misclassification, and reduce by the sample weights of correctly classifying.The updated sample set of weight is used to train next sorter.
It should be noted that, the process extracting the process of histograms of oriented gradients feature of sample image and the histograms of oriented gradients feature of extraction presented hereinbefore image to be identified is similar, does not repeat them here.
In the present embodiment, the histograms of oriented gradients feature of flow process to sample image is trained to carry out classification based training based on adaboost in advance, obtain detection speed adaboost sorter faster, due to using histograms of oriented gradients feature as adaboost training in weak feature, both make use of adaboost feature fast, merge again the feature of the reasonable description contour feature of histograms of oriented gradients, therefore the histograms of oriented gradients feature of image to be identified is extracted, call the histograms of oriented gradients feature that adaboost sorter treats recognition image to identify, achieve fast, high-precision image recognition.
It should be noted that, disclosure embodiment is to for identifying which kind of object in image to be identified does not limit.In a kind of possible embodiment, consider that histograms of oriented gradients feature is particularly suited for describing the feature of human body contour outline feature, disclosure embodiment may be used for identifying human body.Correspondingly, sample image can comprise human sample image and non-human sample image.The adaboost sorter that training obtains can for the adaboost sorter for human detection.Recognition result is the human bioequivalence result of image to be identified.Below in conjunction with Fig. 3, this embodiment is described in detail.Further for example, sample image can comprise face sample image and non-face sample image.The adaboost sorter that training obtains can for the adaboost sorter for Face datection.
Fig. 3 is the process flow diagram of the image-recognizing method according to another exemplary embodiment.The method can be applied to picture pick-up device 110 shown in Fig. 1 or server 120.As shown in Figure 3, the method can comprise:
In step 300, the histograms of oriented gradients feature of human sample image and non-human sample image is extracted.
Such as, the human sample image of magnanimity and non-human sample image can be prepared.The pixel size of these sample images is all normalized to 64*20, obtains two class samples, a class is human sample, and a class is non-human sample.
In step 301, train flow process based on adaboost, classification based training is carried out to the histograms of oriented gradients feature of described human sample image and non-human sample image, obtains the adaboost sorter for human detection.
In the step 310, the histograms of oriented gradients feature of image to be identified is extracted.
In step 320, call the histograms of oriented gradients feature of described adaboost sorter to described image to be identified and identify, obtain the human bioequivalence result of described image to be identified.
Because disclosure embodiment trains flow process to carry out classification based training based on adaboost, obtain detection speed adaboost sorter faster, and, based on histograms of oriented gradients feature, feature interpretation is carried out to human sample image and non-human sample image, both make use of adaboost feature fast, merge again the feature of the reasonable description contour feature of histograms of oriented gradients, call the histograms of oriented gradients feature that adaboost sorter treats recognition image to identify, fast to human bioequivalence speed, that accuracy of identification is high needs can be met.
In order to can the feature of Description Image more accurately, in a kind of possible embodiment, also linear discriminant analysis mapping be carried out to histograms of oriented gradients feature.Below, composition graphs 4 is described in detail to this embodiment.
Fig. 4 is the process flow diagram of the image-recognizing method according to another exemplary embodiment.The method can be applied to picture pick-up device 110 shown in Fig. 1 or server 120.As shown in Figure 4, the method can comprise:
In step 400, the histograms of oriented gradients feature of sample image is extracted.
In step 401, linear discriminant analysis mapping is carried out to the histograms of oriented gradients feature of described sample image, obtain the histograms of oriented gradients feature that described sample image maps through linear discriminant analysis.
In step 402, train flow process based on adaboost, classification based training is carried out to the histograms of oriented gradients feature that described sample image maps through linear discriminant analysis, obtains adaboost sorter.
In step 410, the histograms of oriented gradients feature of image to be identified is extracted.
In step 411, linear discriminant analysis mapping is carried out to the histograms of oriented gradients feature of described image to be identified, obtain the histograms of oriented gradients feature that described image to be identified maps through linear discriminant analysis.
At step 420 which, call described adaboost sorter and the histograms of oriented gradients feature that described image to be identified maps through linear discriminant analysis is identified, obtain recognition result.
Wherein, linear discriminant analysis (LinearDiscriminantAnalysis, LDA) maps, and being also called FLD (FisherLinearDiscriminant, Fisher linear discriminant), is a kind of algorithm for pattern recognition.The basic thought that linear discriminant analysis maps is that the pattern sample of higher-dimension is projected to best discriminant technique vector space, to reach the effect extracting classified information and compressive features space dimensionality, after projection, Assured Mode sample has maximum between class distance and minimum inter-object distance in new subspace, and namely pattern has best separability within this space.Therefore, the present embodiment, by carrying out linear discriminant analysis mapping to histograms of oriented gradients feature, can make the histograms of oriented gradients feature of renewal that obtains after mapping can the contour feature of Description Image more exactly, thus improves accuracy of identification.
In addition, the technical scheme that disclosure embodiment provides can be identified the image that actual scene comprises multiple identification object by the mode of pyramid level traverse scanning.Below, composition graphs 5 is described in detail to this embodiment.
Fig. 5 is the process flow diagram of the image-recognizing method Gen Ju an exemplary embodiment again.The method can be applied to picture pick-up device 110 shown in Fig. 1 or server 120.As shown in Figure 5, the method can comprise:
In step 500, the histograms of oriented gradients feature of sample image is extracted.
In step 501, train flow process based on adaboost, classification based training is carried out to the histograms of oriented gradients feature of described sample image, obtains adaboost sorter.
In step 502, original image is obtained.
Such as, in application scenarios shown in Fig. 1, server 120 can obtain the shooting image of actual scene from picture pick-up device 110, and the model of place of the described actual scene that utilization is set up in advance removes the background in described shooting image, obtains described original image.Because this embodiment can remove the background of shooting image to obtain original image, therefore, the technical scheme that disclosure embodiment provides can identify the image that actual scene comprises background, improves accuracy of identification.
Wherein, the model of place of described actual scene, can utilize ripe GMM Gaussian modeling technology to set up.Wherein, in GMM Gaussian modeling technology, use K Gauss model to carry out the feature of each pixel in token image, after a new two field picture obtains, upgrade mixed Gauss model.Mate with mixed Gauss model with each pixel in present image, if success, judge that this point is as background dot, otherwise be foreground point.The implementation of the background removed in described shooting image at the model of place utilizing ripe GMM Gaussian modeling technology to set up such as can comprise: the model of place utilizing GMM Gaussian modeling technology to set up detects the region that in described shooting image, gray scale changes, the region that these gray scales change is the region of described original image, thus the background in described shooting image is removed in the region changed by extracting described gray scale, obtains described original image.
In step 503, by multiple scaling, convergent-divergent is carried out to described original image, obtain multiple zoomed images.
It should be noted that described scaling can comprise the ratio for amplifying, also can comprising the ratio for reducing, the ratio of geometric ratio can also be comprised, specifically implement to need to arrange according to reality.Such as, original image 601 as shown in Figure 6 in 1/6th, 2/6ths, 3/6ths, 4/6ths, 5/6ths, the ratio of geometric ratio carries out convergent-divergent, can obtain multiple images.Such as, the zoomed image after wherein reducing by 1/6th is image 602.
In step 504, multiple zoomed images described are carried out to the division of identified region, obtain described multiple zoomed images each identified region separately, each identified region is an image to be identified.
Such as, can for often opening zoomed image, the subwindow of exhaustive wherein all positions pre-set dimension is as identified region.Such as, pre-set dimension can be 20*20.
In step 510, the histograms of oriented gradients feature of described identified region is extracted.
Such as, linear discriminant analysis mapping can also be carried out to the histograms of oriented gradients feature of identified region, obtain the histograms of oriented gradients feature that this identified region maps through linear discriminant analysis.
In step 520, call the histograms of oriented gradients feature of described adaboost sorter to each identified region and identify, obtain the recognition result of each identified region.
Such as, the histograms of oriented gradients feature through linear discriminant analysis mapping of described adaboost sorter to each identified region can be called and identify, obtain the recognition result of each identified region.
In step 530, according to each identified region place zoomed image relative to the scaling of original image, each identified region is adjusted to the size relative to original image with same zoom ratio.
Such as, according to the scaling of each identified region place zoomed image relative to original image, each identified region can be returned to the original size relative to described original image geometric ratio.For example, image 602 as shown in Figure 6, the scale down adopted when this image 602 reduces is 1/6th.To the identified region of this image 602 is recovered, then identified region can be amplified 6 times, thus return to the original size corresponding to described original image.
In step 531, by each identified region after adjustment is compared, find out between different identified region, meet the identified region of coincidence condition.
Wherein, described coincidence condition can rule of thumb set, and the disclosure does not limit this.Such as, the identified region of overlapping area more than 20% can be considered as the identified region of satisfied coincidence condition.
In step 532, the recognition result of the identified region of satisfied coincidence condition is merged, obtain the final recognition result of each identified region.
In the present embodiment, have employed and by multiple scaling, convergent-divergent is carried out to original image, Region dividing is carried out to the image after convergent-divergent, the region marked off is identified, a kind of like this recognition method of pyramid level traverse scanning, make the different identified regions of original image can the mating with sample image of maximum probability, thus can the image that actual scene comprises multiple identification object be identified, improve accuracy of identification.
In the disclosure one embodiment, when described recognition result is the human bioequivalence result of described image to be identified, by adding up the final recognition result of each identified region described, the people flow rate statistical result in described actual scene can also be obtained.Below, composition graphs 7 is described in detail to a kind of possible embodiment of this embodiment.
Fig. 7 is the process flow diagram of the image-recognizing method according to another exemplary embodiment.The method can be applied to picture pick-up device 110 shown in Fig. 1 or server 120.As shown in Figure 7, the method can comprise:
In step 700, extract the histograms of oriented gradients feature of human sample image and non-human sample image.
In step 701, linear discriminant analysis mapping is carried out to the histograms of oriented gradients feature of described human sample image and non-human sample image, obtains the histograms of oriented gradients feature mapped through linear discriminant analysis of human sample image and non-human sample image.
In a step 702, train flow process based on adaboost, classification based training is carried out to the histograms of oriented gradients feature through linear discriminant analysis mapping of described human sample image and non-human sample image, obtains the adaboost sorter for human detection.
In step 703, obtain the shooting image of actual scene from picture pick-up device, the model of place of the described actual scene that utilization is set up in advance removes the background in described shooting image, obtains described original image.
In step 704, by multiple scaling, convergent-divergent is carried out to described original image, obtain multiple zoomed images.
In step 705, multiple zoomed images described are carried out to the division of identified region, obtain described multiple zoomed images each identified region separately, each identified region is an image to be identified.
In step 720, the histograms of oriented gradients feature of each identified region is extracted.
In step 711, linear discriminant analysis mapping is carried out to the histograms of oriented gradients feature of each identified region, obtain the histograms of oriented gradients feature that each identified region maps through linear discriminant analysis.
In step 720, call described adaboost sorter and the histograms of oriented gradients feature that each identified region maps through linear discriminant analysis is identified, obtain the human bioequivalence result of each identified region.
In step 730, according to the scaling of each identified region place zoomed image relative to original image, each identified region is returned to the original size relative to described original image geometric ratio.
In step 731, by each identified region returning to original size is compared, find out between different identified region, meet the identified region of coincidence condition.
In step 732, the recognition result of the identified region of satisfied coincidence condition is merged, obtain the final recognition result of each identified region.
Such as, suppose that a certain identified region meets with another identified region the condition of coincidence, then these two identified regions can merge into an identified region, corresponding same final recognition result, such as, this final recognition result can for being judged to be human body, or, be judged to be non-human.
In step 733, the final recognition result of each identified region described is added up, obtain the people flow rate statistical result in described actual scene.
Such as, after supposing to merge, have 10 identified regions, the final recognition result that wherein 5 identified regions are corresponding is for being judged to be human body, and 5 identified regions are judged to be non-human, then can add up the people flow rate statistical result obtained in this actual scene is 5.
In a kind of possible embodiment, corresponding system can also be exported to, such as, for preserving any server or the terminal of this people flow rate statistical result by adding up the people flow rate statistical result obtained.
In the present embodiment, owing to carrying out convergent-divergent to original image by multiple scaling, Region dividing is carried out to the image after convergent-divergent, to the histograms of oriented gradients feature in the region marked off, the adaboost sorter being used for human detection is utilized to identify, therefore, it is possible to carry out human bioequivalence to the image that actual scene comprises multiple human body fast, accurately, obtain the people flow rate statistical result in actual scene.
Fig. 8 is the block diagram of the pattern recognition device according to an exemplary embodiment.This device can be configured at picture pick-up device 110 shown in Fig. 1 or server 120.As shown in Figure 8, this device can comprise: characteristic extracting module 810, identification module 820.
This characteristic extracting module 810, can be configured to the histograms of oriented gradients feature extracting image to be identified.
This identification module 820, can be configured to call the histograms of oriented gradients feature of adaboost sorter to the image described to be identified that described characteristic extracting module 810 is extracted identify, obtain recognition result, wherein, described adaboost sorter trains the histograms of oriented gradients feature of flow process to sample image to carry out classification based training acquisition based on adaboost in advance.
Wherein, histograms of oriented gradients (HistogramofOrientedGradient, hog) feature is a kind of feature interpretation operator being used for carrying out object detection in computer vision and image procossing.It carrys out constitutive characteristic by the gradient orientation histogram of calculating and statistical picture regional area.
Such as, the process extracting the histograms of oriented gradients feature of image to be identified comprises: treat recognition image and be normalized, calculate the histogram of gradients of normalized sample image, each cell block is carried out to the projection of default weight to histogram of gradients, contrast normalization is carried out for the cell block in each overlapping block, histogram vectors in all overlapping blocks is combined as histograms of oriented gradients proper vector, is histograms of oriented gradients feature.
Wherein, adaboost (AdaptiveBoosting, self-adaptation strengthens) is a kind of machine learning method.The training flow process of this machine learning method is a kind of iterative algorithm, takes turns in iteration add a new Weak Classifier, until reach certain predetermined enough little error rate at each.Training flow process comprises: each training sample is endowed a weight, show that it is selected into the probability of training set by certain sorter, if certain sample point is classified exactly, so under construction in a training set, its selected probability is just lowered, on the contrary, if certain sample point is not classified exactly, so its weight is just improved.By such training flow process, adaboost can focus on more indistinguishable sample.Such as, the weight of each sample can be made at first equal, for kth time iterative operation, choose sample point according to weight, and then training classifier C
k.Then according to sorter C
k, improve by the weight of the sample of its misclassification, and reduce by the sample weights of correctly classifying.The updated sample set of weight is used to train next sorter.
It should be noted that, the process extracting the process of histograms of oriented gradients feature of sample image and the histograms of oriented gradients feature of extraction presented hereinbefore image to be identified is similar, does not repeat them here.
In the present embodiment, the histograms of oriented gradients feature of flow process to sample image is trained to carry out classification based training based on adaboost in advance, obtain detection speed adaboost sorter faster, due to using histograms of oriented gradients feature as adaboost training in weak feature, both make use of adaboost feature fast, merge again the feature of the reasonable description contour feature of histograms of oriented gradients, therefore the histograms of oriented gradients feature of image to be identified is extracted, call the histograms of oriented gradients feature that adaboost sorter treats recognition image to identify, achieve fast, high-precision image recognition.
It should be noted that, disclosure embodiment is to for identifying which kind of object in image to be identified does not limit.In a kind of possible embodiment, described sample image can comprise: human sample image and non-human sample image.Correspondingly, described adaboost sorter is the adaboost sorter for human detection.Described recognition result is the human bioequivalence result of described image to be identified.Such as, the human sample image of magnanimity and non-human sample image can be prepared.The pixel size of these sample images is all normalized to 64*20, obtains two class samples, a class is human sample, and a class is non-human sample.Because disclosure embodiment trains flow process to carry out classification based training based on adaboost, obtain detection speed adaboost sorter faster, and, based on histograms of oriented gradients feature, feature interpretation is carried out to human sample image and non-human sample image, both make use of adaboost feature fast, merge again the feature of the reasonable description contour feature of histograms of oriented gradients, call the histograms of oriented gradients feature that adaboost sorter treats recognition image to identify, fast to human bioequivalence speed, that accuracy of identification is high needs can be met.
In order to can the feature of Description Image more accurately, in a kind of possible embodiment, also linear discriminant analysis mapping be carried out to histograms of oriented gradients feature.Wherein, the histograms of oriented gradients of described sample image is characterized as the histograms of oriented gradients feature mapped through linear discriminant analysis.Described characteristic extracting module 810 can also be configured to carry out linear discriminant analysis mapping to the histograms of oriented gradients feature of described image to be identified, obtains the histograms of oriented gradients feature that described image to be identified maps through linear discriminant analysis.Described identification module 820 can be configured to call adaboost sorter and identify the histograms of oriented gradients feature through linear discriminant analysis mapping that described characteristic extracting module is extracted, and obtains recognition result.
Wherein, linear discriminant analysis (LinearDiscriminantAnalysis, LDA) maps, and being also called FLD (FisherLinearDiscriminant, Fisher linear discriminant), is a kind of algorithm for pattern recognition.The basic thought that linear discriminant analysis maps is that the pattern sample of higher-dimension is projected to best discriminant technique vector space, to reach the effect extracting classified information and compressive features space dimensionality, after projection, Assured Mode sample has maximum between class distance and minimum inter-object distance in new subspace, and namely pattern has best separability within this space.Therefore, the present embodiment, by carrying out linear discriminant analysis mapping to histograms of oriented gradients feature, can make the histograms of oriented gradients feature of renewal that obtains after mapping can the contour feature of Description Image more exactly, thus improves accuracy of identification.
In addition, the technical scheme that disclosure embodiment provides can be identified the image that actual scene comprises multiple identification object by the mode of pyramid level traverse scanning.Below, composition graphs 9 is described in detail to this embodiment.
Fig. 9 is the block diagram of the pattern recognition device according to another exemplary embodiment.This device can be configured at picture pick-up device 110 shown in Fig. 1 or server 120.As shown in Figure 9, this device can also comprise: original image acquisition module 800, Zoom module 801, Region dividing module 802, size adjustment module 830, comparison module 831 and merging module 832.
This original image acquisition module 800, can be configured to obtain original image.
This Zoom module 801, the original image that can be configured to described original image acquisition module 800 obtains carries out convergent-divergent by multiple scaling, obtains multiple zoomed images.
This Region dividing module 802, can be configured to the division multiple zoomed images that described Zoom module 801 obtains being carried out to identified region, and obtain described multiple zoomed images each identified region separately, each identified region is an image to be identified.
Wherein, described identification module 820 is configured to call the histograms of oriented gradients feature of described adaboost sorter to each identified region that described Region dividing module 802 obtains and identifies, obtains the recognition result of each identified region.
This size adjustment module 830, can be configured to according to each identified region place zoomed image, relative to the scaling of original image, each identified region be adjusted to the size relative to original image with same zoom ratio;
This comparison module 831, is configured to be compared by each identified region after described size adjustment module 830 being adjusted, finds out between different identified region, meet the identified region of coincidence condition;
This merging module 832, is configured to the recognition result of the identified region of satisfied coincidence condition to merge, obtains the final recognition result of each identified region.
In the present embodiment, owing to carrying out convergent-divergent to original image by multiple scaling, Region dividing is carried out to the image after convergent-divergent, to the histograms of oriented gradients feature in the region marked off, the adaboost sorter being used for human detection is utilized to identify, therefore, it is possible to carry out human bioequivalence to the image that actual scene comprises multiple human body fast, accurately, obtain the people flow rate statistical result in actual scene.
Such as, application scenarios according to Fig. 1, as shown in Figure 9, described original image acquisition module 800 can comprise: shooting Image Acquisition submodule 8001 and background detection submodule 8002.
This shooting Image Acquisition submodule 8001, can be configured to the shooting image obtaining actual scene from picture pick-up device.
This background detection submodule 8002, can be configured to utilize the model of place of the described actual scene set up in advance to remove background in described shooting image, obtain described original image.
In the disclosure one embodiment, when described recognition result is the human bioequivalence result of described image to be identified, by adding up the final recognition result of each identified region described, the people flow rate statistical result in described actual scene can also be obtained.In this embodiment, as shown in Figure 9, this device can also comprise: statistical module 840, can be configured to when described recognition result is the human bioequivalence result of described image to be identified, the final recognition result of each identified region described is added up, obtains the people flow rate statistical result in described actual scene.
In the present embodiment, owing to carrying out convergent-divergent to original image by multiple scaling, Region dividing is carried out to the image after convergent-divergent, to the histograms of oriented gradients feature in the region marked off, the adaboost sorter being used for human detection is utilized to identify, therefore, it is possible to carry out human bioequivalence to the image that actual scene comprises multiple human body fast, accurately, obtain the people flow rate statistical result in actual scene.
About the device in above-described embodiment, wherein the concrete mode of modules executable operations has been described in detail in about the embodiment of the method, will not elaborate explanation herein.
Figure 10 is the block diagram of a kind of pattern recognition device 1000 according to an exemplary embodiment.Such as, device 1000 may be provided in a server.With reference to Figure 10, device 1000 comprises processing components 1022, and it comprises one or more processor further, and the memory resource representated by storer 1032, can such as, by the instruction of the execution of processing components 1022, application program for storing.The application program stored in storer 1032 can comprise each module corresponding to one group of instruction one or more.In addition, processing components 1022 is configured to perform instruction, to perform above-mentioned image-recognizing method.
Device 1000 can also comprise the power management that a power supply module 1026 is configured to actuating unit 1000, and a wired or wireless network interface 1050 is configured to device 1000 to be connected to network, and input and output (I/O) interface 1058.Device 1000 can operate the operating system based on being stored in storer 1032, such as WindowsServerTM, MacOSXTM, UnixTM, LinuxTM, FreeBSDTM or similar.
Those skilled in the art, at consideration instructions and after putting into practice the disclosure, will easily expect other embodiment of the present disclosure.The application is intended to contain any modification of the present disclosure, purposes or adaptations, and these modification, purposes or adaptations are followed general principle of the present disclosure and comprised the undocumented common practise in the art of the disclosure or conventional techniques means.Instructions and embodiment are only regarded as exemplary, and true scope of the present disclosure and spirit are pointed out by claim below.
Should be understood that, the disclosure is not limited to precision architecture described above and illustrated in the accompanying drawings, and can carry out various amendment and change not departing from its scope.The scope of the present disclosure is only limited by appended claim.
Claims (13)
1. an image-recognizing method, is characterized in that, comprising:
Extract the histograms of oriented gradients feature of image to be identified;
Call the histograms of oriented gradients feature of adaboost sorter to described image to be identified to identify, obtain recognition result, wherein, described adaboost sorter trains the histograms of oriented gradients feature of flow process to sample image to carry out classification based training acquisition based on adaboost in advance.
2. method according to claim 1, is characterized in that, described sample image comprises: human sample image and non-human sample image;
Described adaboost sorter is the adaboost sorter for human detection;
Described recognition result is the human bioequivalence result of described image to be identified.
3. method according to claim 1 and 2, is characterized in that, the histograms of oriented gradients of described sample image is characterized as the histograms of oriented gradients feature mapped through linear discriminant analysis;
Described method also comprises: carry out linear discriminant analysis mapping to the histograms of oriented gradients feature of described image to be identified, obtains the histograms of oriented gradients feature that described image to be identified maps through linear discriminant analysis;
The histograms of oriented gradients feature of the described adaboost of calling sorter to described image to be identified identifies, obtain recognition result to comprise: call described adaboost sorter and identify the histograms of oriented gradients feature that described image to be identified maps through linear discriminant analysis, obtain recognition result.
4. method according to claim 1 and 2, is characterized in that, described method also comprises: obtain original image; By multiple scaling, convergent-divergent is carried out to described original image, obtains multiple zoomed images; Multiple zoomed images described are carried out to the division of identified region, obtain described multiple zoomed images each identified region separately, each identified region is an image to be identified;
Wherein, the histograms of oriented gradients feature of the described adaboost of calling sorter to described image to be identified identifies, obtain recognition result to comprise: call the histograms of oriented gradients feature of described adaboost sorter to each identified region and identify, obtain the recognition result of each identified region;
Described method also comprises: according to the scaling of each identified region place zoomed image relative to original image, each identified region is adjusted to the size relative to original image with same zoom ratio, by each identified region after adjustment is compared, find out between different identified region, meet the identified region of coincidence condition, the recognition result of the identified region of satisfied coincidence condition is merged, obtains the final recognition result of each identified region.
5. method according to claim 4, is characterized in that, described acquisition original image comprises:
The shooting image of actual scene is obtained from picture pick-up device;
The model of place of the described actual scene that utilization is set up in advance removes the background in described shooting image, obtains described original image.
6. method according to claim 4, is characterized in that, when described recognition result is the human bioequivalence result of described image to be identified, described method also comprises:
The final recognition result of each identified region described is added up, obtains the people flow rate statistical result in described actual scene.
7. a pattern recognition device, is characterized in that, comprising:
Characteristic extracting module, is configured to the histograms of oriented gradients feature extracting image to be identified;
Identification module, be configured to call the histograms of oriented gradients feature of adaboost sorter to the image described to be identified that described characteristic extracting module is extracted identify, obtain recognition result, wherein, described adaboost sorter trains the histograms of oriented gradients feature of flow process to sample image to carry out classification based training acquisition based on adaboost in advance.
8. device according to claim 7, is characterized in that, described sample image comprises: human sample image and non-human sample image;
Described adaboost sorter is the adaboost sorter for human detection;
Described recognition result is the human bioequivalence result of described image to be identified.
9. device according to claim 1 and 2, is characterized in that, the histograms of oriented gradients of described sample image is characterized as the histograms of oriented gradients feature mapped through linear discriminant analysis;
Described characteristic extracting module is also configured to, and carries out linear discriminant analysis mapping to the histograms of oriented gradients feature of described image to be identified, obtains the histograms of oriented gradients feature that described image to be identified maps through linear discriminant analysis;
Described identification module is configured to, and calls adaboost sorter and identifies the histograms of oriented gradients feature through linear discriminant analysis mapping that described characteristic extracting module is extracted, obtain recognition result.
10. device according to claim 1 and 2, is characterized in that, described device also comprises:
Original image acquisition module, is configured to obtain original image;
Zoom module, the original image be configured to described original image acquisition module obtains carries out convergent-divergent by multiple scaling, obtains multiple zoomed images;
Region dividing module, is configured to the division multiple zoomed images that described Zoom module obtains being carried out to identified region, and obtain described multiple zoomed images each identified region separately, each identified region is an image to be identified;
Wherein, described identification module is configured to call the histograms of oriented gradients feature of described adaboost sorter to each identified region that described Region dividing module obtains and identifies, obtains the recognition result of each identified region;
Described device also comprises:
Size adjustment module, to be configured to according to each identified region place zoomed image, relative to the scaling of original image, each identified region be adjusted to the size relative to original image with same zoom ratio;
Comparison module, is configured to, by being compared by each identified region after described size adjustment module adjustment, find out between different identified region, meet the identified region of coincidence condition;
Merge module, be configured to the recognition result of the identified region of satisfied coincidence condition to merge, obtain the final recognition result of each identified region.
11. devices according to claim 10, is characterized in that, described original image acquisition module comprises:
Shooting Image Acquisition submodule, is configured to the shooting image obtaining actual scene from picture pick-up device;
Background detection submodule, is configured to utilize the model of place of the described actual scene set up in advance to remove background in described shooting image, obtains described original image.
12. devices according to claim 10, is characterized in that, described device also comprises:
Statistical module, is configured to, when described recognition result is the human bioequivalence result of described image to be identified, adds up, obtain the people flow rate statistical result in described actual scene to the final recognition result of each identified region described.
13. 1 kinds of pattern recognition devices, is characterized in that, comprising:
Processor;
For the storer of storage of processor executable instruction;
Wherein, described processor is configured to:
Extract the histograms of oriented gradients feature of image to be identified;
Call the histograms of oriented gradients feature of adaboost sorter to described image to be identified to identify, obtain recognition result, wherein, described adaboost sorter trains the histograms of oriented gradients feature of flow process to sample image to carry out classification based training acquisition based on adaboost in advance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510881619.2A CN105512664A (en) | 2015-12-03 | 2015-12-03 | Image recognition method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510881619.2A CN105512664A (en) | 2015-12-03 | 2015-12-03 | Image recognition method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105512664A true CN105512664A (en) | 2016-04-20 |
Family
ID=55720631
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510881619.2A Pending CN105512664A (en) | 2015-12-03 | 2015-12-03 | Image recognition method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105512664A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106557768A (en) * | 2016-11-25 | 2017-04-05 | 北京小米移动软件有限公司 | The method and device is identified by word in picture |
| CN107978185A (en) * | 2017-12-07 | 2018-05-01 | 何旭连 | A kind of good children learning machine of teaching efficiency |
| CN112712066A (en) * | 2021-01-19 | 2021-04-27 | 腾讯科技（深圳）有限公司 | Image recognition method and device, computer equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100202707A1 (en) * | 2004-12-29 | 2010-08-12 | Fotonation Vision Limited | Method and Component for Image Recognition |
| CN101859436A (en) * | 2010-06-09 | 2010-10-13 | 王巍 | Large-amplitude regular movement background intelligent analysis and control system |
| CN102147869A (en) * | 2011-03-31 | 2011-08-10 | 上海交通大学 | Pedestrian detection method based on foreground analysis and pattern recognition |
| CN102682287A (en) * | 2012-04-17 | 2012-09-19 | 电子科技大学 | Pedestrian detection method based on saliency information |
-
2015
- 2015-12-03 CN CN201510881619.2A patent/CN105512664A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100202707A1 (en) * | 2004-12-29 | 2010-08-12 | Fotonation Vision Limited | Method and Component for Image Recognition |
| CN101859436A (en) * | 2010-06-09 | 2010-10-13 | 王巍 | Large-amplitude regular movement background intelligent analysis and control system |
| CN102147869A (en) * | 2011-03-31 | 2011-08-10 | 上海交通大学 | Pedestrian detection method based on foreground analysis and pattern recognition |
| CN102682287A (en) * | 2012-04-17 | 2012-09-19 | 电子科技大学 | Pedestrian detection method based on saliency information |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106557768A (en) * | 2016-11-25 | 2017-04-05 | 北京小米移动软件有限公司 | The method and device is identified by word in picture |
| CN106557768B (en) * | 2016-11-25 | 2021-07-06 | 北京小米移动软件有限公司 | Method and device for recognizing characters in picture |
| CN107978185A (en) * | 2017-12-07 | 2018-05-01 | 何旭连 | A kind of good children learning machine of teaching efficiency |
| CN107978185B (en) * | 2017-12-07 | 2019-12-03 | 南京白下高新技术产业园区投资发展有限责任公司 | A kind of good children learning machine of teaching efficiency |
| CN112712066A (en) * | 2021-01-19 | 2021-04-27 | 腾讯科技（深圳）有限公司 | Image recognition method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11062123B2 (en) | Method, terminal, and storage medium for tracking facial critical area | |
| CN110738207B (en) | Character detection method for fusing character area edge information in character image | |
| CN109359538B (en) | Training method of convolutional neural network, gesture recognition method, device and equipment | |
| KR102591961B1 (en) | Model training method and device, and terminal and storage medium for the same | |
| CN107679448B (en) | Eyeball action-analysing method, device and storage medium | |
| CN109960742B (en) | Local information searching method and device | |
| CN108961315B (en) | Target tracking method and device, computer equipment and storage medium | |
| KR101896357B1 (en) | Method, device and program for detecting an object | |
| CN102422325B (en) | Pattern recognition apparatus and method therefor configured to recognize object and another lower-order object | |
| CN111523414A (en) | Face recognition method and device, computer equipment and storage medium | |
| CN105512683A (en) | Target positioning method and device based on convolution neural network | |
| US9443137B2 (en) | Apparatus and method for detecting body parts | |
| CN111160275B (en) | Pedestrian re-recognition model training method, device, computer equipment and storage medium | |
| CN110751232A (en) | Chinese complex scene text detection and identification method | |
| CN112381092B (en) | Tracking method, tracking device and computer readable storage medium | |
| CN103105924A (en) | Man-machine interaction method and device | |
| CN112668374A (en) | Image processing method and device, re-recognition network training method and electronic equipment | |
| CN112241667A (en) | Image detection method, device, equipment and storage medium | |
| CN113706481A (en) | Sperm quality detection method, sperm quality detection device, computer equipment and storage medium | |
| CN111046746A (en) | License plate detection method and device | |
| CN115187844A (en) | Image identification method and device based on neural network model and terminal equipment | |
| CN113780145A (en) | Sperm morphology detection method, sperm morphology detection device, computer equipment and storage medium | |
| Ahmadi et al. | Efficient and fast objects detection technique for intelligent video surveillance using transfer learning and fine-tuning | |
| CN105512664A (en) | Image recognition method and device | |
| CN116958873A (en) | Pedestrian tracking method, device, electronic equipment and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160420 |