CN108874889B - Target body retrieval method, system and device based on target body image - Google Patents
Target body retrieval method, system and device based on target body image Download PDFInfo
- Publication number
- CN108874889B CN108874889B CN201810460265.8A CN201810460265A CN108874889B CN 108874889 B CN108874889 B CN 108874889B CN 201810460265 A CN201810460265 A CN 201810460265A CN 108874889 B CN108874889 B CN 108874889B
- Authority
- CN
- China
- Prior art keywords
- image
- target volume
- training
- model
- retrieval
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000012549 training Methods 0.000 claims abstract description 64
- 238000012545 processing Methods 0.000 claims abstract description 16
- 238000004364 calculation method Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 36
- 238000005457 optimization Methods 0.000 claims description 24
- 238000010801 machine learning Methods 0.000 claims description 17
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 230000009977 dual effect Effects 0.000 claims description 4
- 238000012847 principal component analysis method Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 abstract 1
- 238000012360 testing method Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000011423 initialization method Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the technical field of image processing, in particular provides a target body retrieval method, a target body retrieval system and a target body retrieval device based on a target body image, and aims to solve the technical problem of improving the image retrieval accuracy. For this purpose, the target body retrieval method in the invention extracts the image features of the target body image based on the pre-constructed image retrieval model, calculates the similarity between the image features of the target body image and the image features of the comparison images in the preset image library, and obtains the comparison images similar to the target body image in the image library according to the calculation result, thereby determining the image category of the target body image according to the image category of the similarity comparison images. The image retrieval model is subjected to model training by using a two-step training method (firstly, a classification loss function is adopted for training, and then, a dual-threshold loss function is adopted for training), so that the accuracy of the retrieval result of the image retrieval model can be improved. Meanwhile, the device and the system can realize the target body retrieval method.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a target body retrieval method, a system and a device based on a target body image.
Background
The image retrieval technology refers to an image processing technology for acquiring similar images based on text, and/or content, and/or semantic information of a target image. Currently, image retrieval can be performed using image processing techniques based on machine learning algorithms. For example, using a deep convolutional neural network, image features of a target image are acquired, and similar images are retrieved based on the acquired image features. In this process, the deep convolutional neural network is usually trained by using a contrast loss function based on a single threshold. However, the training mode only sets a threshold for the dissimilar image pair, and it is difficult to ensure that the contribution proportion of the similar image and the dissimilar image to the loss function is the same, that is, the training process focuses on the similar image pair, so that the generalization capability of the deep convolutional neural network is reduced, and the similar image of the unknown class of image cannot be accurately retrieved.
Disclosure of Invention
The method aims to solve the problems in the prior art, namely, the technical problem of how to improve the image retrieval accuracy is solved. To this end, the invention provides a target body retrieval method, a system and a device based on a target body image.
In a first aspect, a method for retrieving a target volume based on a target volume image in the present invention includes the following steps:
extracting image features of the target volume image based on a pre-constructed image retrieval model;
calculating the similarity between the image characteristics of the target image and the image characteristics of the comparison images in a preset image library;
according to the calculation result, a comparison image similar to the target body image in the image library is obtained;
the image retrieval model is a model which is based on a preset image training sample and is constructed by utilizing a machine learning algorithm.
Further, a preferred technical solution provided by the present invention is:
before the step of calculating the similarity between the image features of the target volume image and the image features of the comparison images in the preset image library, the method further comprises the following steps:
performing dimension reduction processing on the image features of the target volume image according to the image feature dimensions of the compared images in the image library;
in this case, the step of "calculating the similarity between the image feature of the target volume image and the image feature of the comparison image in the preset image library" further includes calculating the similarity between the image feature of the target volume image after the dimension reduction processing and the image feature of the comparison image.
Further, a preferred technical solution provided by the present invention is:
the step of performing dimension reduction processing on the image features of the target volume image includes performing dimension reduction processing on the image features of the target volume image by using a principal component analysis method.
Further, a preferred technical solution provided by the present invention is:
the image retrieval model is a model based on a Siamese network framework;
before the step of extracting the image features of the target volume image based on the pre-constructed image retrieval model, the method further comprises:
acquiring image characteristics of the image training sample;
based on a preset classification loss function and the image characteristics of the image training sample, performing model training on the image retrieval model by using a machine learning algorithm to obtain a primary optimization model;
and performing model training on the preliminary optimization model by using a machine learning algorithm based on a preset double-threshold loss function and the image characteristics of the image training samples to obtain a final optimization model.
Further, a preferred technical solution provided by the present invention is:
before the step of performing model training on the preliminary optimization model by using a machine learning algorithm based on the preset dual-threshold loss function and the image features of the image training samples to obtain a final optimization model, the method further includes:
acquiring a similar image sample pair and a dissimilar image sample pair according to the image training sample;
respectively calculating the probability distribution of the characteristic distances of the similar image sample pairs and the probability distribution of the characteristic distances of the dissimilar image sample pairs;
and obtaining a threshold corresponding to the similar image sample pair and a threshold corresponding to the dissimilar image sample pair in the double-threshold loss function according to the probability distribution.
Further, a preferred technical solution provided by the present invention is:
before the step of performing model training on the image retrieval model by using a machine learning algorithm based on a preset classification loss function and the image features of the image training samples to obtain a preliminary optimization model, the method further comprises:
acquiring the parameter weight of a neural network which finishes network training in advance, and performing parameter weight initialization on the image retrieval model according to the parameter weight;
the neural network is obtained by performing network training by using the machine learning algorithm based on the ImageNet data set.
Further, a preferred technical solution provided by the present invention is:
the classification loss function is shown as follows:
Lc=-wklog(pk)
wherein, L iscRepresents a classification loss function, said wkWeight corresponding to the kth preset image category, pkThe probability that the image class of a certain image sample is the kth image class.
Further, a preferred technical solution provided by the present invention is:
the dual threshold loss function is shown as follows:
wherein, L (I)p,Iq) Represents a dual threshold loss function, said (I)p,Iq) Representing pairs of image samples, said alpha1Is the threshold value corresponding to the similar image sample pair, the alpha2A threshold corresponding to a dissimilar image sample pair; y is a label value corresponding to the image sample pair, and y is 1 to indicate that the image sample pair is a similar imageA sample pair, y-0 representing that the image sample pair is a dissimilar image sample pair; d | | | f (I)p)-f(Iq)||2The | | | f (I)p)-f(Iq)||2Denotes f (I)p)-f(Iq) 2 norm of, said f (I)p) As an image IpThe image feature of f (I)q) As an image IqThe image feature of (1).
In a second aspect, the storage device in the present invention stores a plurality of programs adapted to be loaded by a processor to execute the target volume retrieval method based on a target volume image according to the above-described technical solution.
In a third aspect, the control apparatus in the present invention comprises a processor and a storage device, the storage device being adapted to store a plurality of programs, the programs being adapted to be loaded by the processor to perform the target volume retrieval method based on a target volume image according to the above-mentioned technical solution.
Compared with the closest prior art, the technical scheme at least has the following beneficial effects:
1. the target body retrieval method based on the target body image mainly comprises the steps of extracting image features of the target body image based on a pre-constructed image retrieval model, calculating the similarity between the image features of the target body image and the image features of comparison images in a preset image library, and obtaining the comparison images similar to the target body image in the image library according to the calculation result, so that the image category of the target body image can be determined according to the image category of the similar comparison images.
2. The target body retrieval method based on the target body image firstly utilizes the classification loss function to carry out model training on the image retrieval model to obtain a preliminary optimization model, and then utilizes the dual-threshold loss function to carry out model training on the preliminary optimization model to obtain a final optimization model. Based on the method, the loss balance of the similar image sample pair and the dissimilar image sample pair in the model training process can be ensured, and the image recognition performance of the image retrieval model is further improved.
Drawings
Fig. 1 is a schematic diagram illustrating the main steps of a target volume retrieval method based on a target volume image according to an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating the main steps of a model training method for an image retrieval model according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of the main structure of an image retrieval model based on the Siemese network framework in the embodiment of the present invention;
FIG. 4 is a diagram illustrating a testing process of an image retrieval model according to an embodiment of the present invention.
Detailed Description
Preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only for explaining the technical principle of the present invention, and are not intended to limit the scope of the present invention.
Referring to fig. 1, fig. 1 schematically illustrates the main steps of a target volume retrieval method based on a target volume image in the present embodiment. As shown in fig. 1, the image retrieval can be performed according to the following steps in this embodiment:
step S101: and extracting the image characteristics of the target volume image based on a pre-constructed image retrieval model.
In this embodiment, the image retrieval model is a model constructed based on a preset image training sample and by using a machine learning algorithm. Optionally, in this embodiment, the image retrieval model may be a neural network model based on a siernese network framework, such as AlexNet, VGGNet, or ResNet. .
Step S102: and calculating the similarity between the image characteristics of the target image and the image characteristics of the comparison image in a preset image library. Specifically, in this embodiment, the similarity between the target volume image and the comparison image may be calculated according to the following steps:
step S1021: and performing dimension reduction processing on the image features of the target image according to the image feature dimensions of the compared images in the image library. Optionally, in this embodiment, a principal component analysis method may be used to perform dimension reduction processing on the image feature of the target volume image, so that the image feature dimension of the target volume image is the same as the image feature dimension of the comparison image. The image feature dimension of the aligned images in the image library in this embodiment may be 64, 128, or 256.
Step S1022: and calculating the similarity between the image characteristics of the target volume image after the dimension reduction processing and the image characteristics of the comparison image. Optionally, in this embodiment, the cosine similarity between the image feature of the target volume image and the image feature of the comparison image may be calculated.
Step S103: and according to the calculation result, obtaining a comparison image similar to the target body image in the image library. Specifically, in this embodiment, the comparison image corresponding to the maximum similarity may be used as the similar image of the target volume image.
Further, in this embodiment, before searching for a similar image of a target image by using the target search method shown in fig. 1, model training may be performed on the image search model to improve the accuracy of the search result. The following describes a model training method of the image search model in this embodiment with reference to the drawings.
Referring to fig. 2, fig. 2 illustrates the main steps of the model training method for the image retrieval model in the present embodiment. As shown in fig. 2, in this embodiment, the image retrieval model may be model-trained according to the following steps:
step S201: and initializing parameters of the image retrieval model.
Specifically, in this embodiment, the parameter initialization may be performed on the image retrieval model according to the following steps: firstly, the parameter weight of the neural network which finishes the network training in advance is obtained. Then, parameter weight initialization is carried out on the image retrieval model according to the parameter weight of the neural network. The network training is completed in advance, namely the neural network is obtained by performing network training by using a machine learning algorithm based on the ImageNet data set.
Step S202: and acquiring image characteristics of the image training sample.
Specifically, in the present embodiment, the image features of the image training sample may be obtained by using an image retrieval model that completes parameter initialization.
Step S203: and performing model training on the image retrieval model by using a machine learning algorithm based on a preset classification loss function and the image characteristics of the image training sample to obtain a primary optimization model.
Specifically, the classification loss function in the present embodiment may be a cross-entropy loss function as shown in the following formula (1):
Lc=-wklog(pk) (1)
the meaning of each parameter in the formula (1) is as follows:
Lcrepresenting a classification loss function, wkWeight, p, corresponding to the kth preset image classkThe probability that the image class of a certain image sample is the kth image class.
Step S204: and obtaining a threshold corresponding to the similar image sample pair and a threshold corresponding to the dissimilar image sample pair in the dual-threshold loss function.
Specifically, in this embodiment, the threshold corresponding to the pair of the similar image sample and the dissimilar image sample may be obtained according to the following steps:
step S2041: and acquiring a similar image sample pair and a dissimilar image sample pair according to the image training sample.
Specifically, in the present embodiment, the pair of similar image samples and the pair of dissimilar image samples may be randomly sampled from the image training samples.
Step S2042: and respectively calculating the probability distribution of the characteristic distances of the similar image sample pairs and the probability distribution of the characteristic distances of the dissimilar image sample pairs.
Specifically, in this embodiment, the euclidean distance between the feature distances of the image sample pairs (similar image sample pairs or dissimilar image sample pairs) may be calculated, and then the probability distribution of the corresponding image sample pairs is obtained according to the euclidean distance.
Step S2043: and obtaining a threshold corresponding to the similar image sample pair and a threshold corresponding to the dissimilar image sample pair in the dual-threshold loss function according to the probability distribution.
Specifically, in this embodiment, an average value or a median of the probability distribution may be selected as an initial threshold value, and a final threshold value may be determined by using an experimental test method.
Step S205: and performing model training on the preliminary optimization model by using a machine learning algorithm based on the preset double-threshold loss function and the image characteristics of the image training sample to obtain a final optimization model. Optionally, in this embodiment, a model training may be performed on the preliminary optimization model by using an error back propagation algorithm.
Specifically, the dual-threshold loss function in the present embodiment may be a loss function as shown in the following equation (2):
the meaning of each parameter in the formula (2) is as follows:
L(Ip,Iq) Representing a dual threshold loss function, (I)p,Iq) Representing pairs of image samples, α1For the corresponding threshold, alpha, of the pair of similar image samples2Y is a label value corresponding to the image sample pair, and y is 1, which represents that the image sample pair is a similar image sample pair, and y is 0, which represents that the image sample pair is a dissimilar image sample pair; d | | | f (I)p)-f(Iq)||2,||f(Ip)-f(Iq)||2Denotes f (I)p)-f(Iq) 2 norm of f (I)p) As an image IpThe image feature of f (I)q) As an image IqThe image feature of (1).
Next, a model training method of the image retrieval model shown in fig. 2 in this embodiment will be described by taking a preset gun image library as an example. The gun image library in this embodiment contains 14755 gun images and these gun images belong to 167 gun categories. Wherein, the image training sample comprises 107 gun classes, the image verification sample comprises 20 gun classes, and the image testing sample comprises 40 gun classes.
Specifically, in this embodiment, the model training method may be performed on the image retrieval model according to the following steps:
step S301: and initializing parameters of the image retrieval model.
The parameter initialization method in this embodiment is the same as the parameter initialization method in the model training method shown in fig. 2, and for brevity of description, no further description is given here.
Step S302: and acquiring image characteristics of the image training sample.
Specifically, in this embodiment, about 40 ten thousand of images are acquired in an image training sample, and an image feature of each acquired image training sample is extracted using an image retrieval model.
Step S303: and performing model training on the image retrieval model by using a cross entropy loss function to obtain a primary optimization model.
Step S304: and respectively calculating the probability distribution of the characteristic distances of the similar image sample pairs and the probability distribution of the characteristic distances of the dissimilar image sample pairs.
Step S305: and obtaining a threshold corresponding to the similar image sample pair and a threshold corresponding to the dissimilar image sample pair in the dual-threshold loss function according to the probability distribution.
Specifically, the threshold corresponding to the similar image sample pair in this embodiment is 0.8, and the threshold corresponding to the dissimilar image sample pair is 1.2.
Step S306: and performing model training on the preliminary optimization model by using a dual-threshold loss function and the image characteristics of the image training samples and using a machine learning algorithm to obtain a final optimization model. In this case, in this embodiment, the final optimization model may also be tested by using the image test samples, so as to obtain the feature vector of each image test sample.
Specifically, the image training sample corresponding to each gun category in the present embodiment includes 180 similar image sample pairs and 180 dissimilar image sample pairs. The iteration cycle in the model training process is 30, and each iteration cycle comprises 602 iterations, and each iteration uses 64 image sample pairs.
Referring to fig. 3, fig. 3 illustrates the main structure of the image retrieval model based on the siemese network framework in the present embodiment. As shown in fig. 3, the image retrieval model in this embodiment is a model based on a dual-branch Siamese network framework, and two image samples of a certain image sample pair are respectively input into two branches of the Siamese network, and are sequentially subjected to convolution operation, feature extraction, normalization processing, and dual-threshold loss function calculation.
Referring to fig. 4, fig. 4 illustrates a process of testing the image retrieval model in the present embodiment. As shown in fig. 4, in the present embodiment, in the test process, the image retrieval model may perform convolution operation, feature extraction, and normalization on the image test sample in sequence to obtain a feature vector of the image test sample.
Therefore, model training and model testing of the image retrieval model are completed, the trained image retrieval model can be used for retrieving the gun images to obtain similar gun images, and the class of the gun is further judged.
Although the foregoing embodiments describe the steps in the above sequential order, those skilled in the art will understand that, in order to achieve the effect of the present embodiments, the steps may not be executed in such an order, and may be executed simultaneously (in parallel) or in an inverse order, and these simple variations are within the scope of the present invention.
Further, based on the above-mentioned embodiment of the target volume retrieval method, the present invention also provides a storage device, in which a plurality of programs are stored, and the programs may be adapted to be loaded by a processor to execute the target volume retrieval method described in the above-mentioned embodiment of the method.
Still further, based on the above-mentioned embodiment of the target volume retrieval method, the present invention also provides a control apparatus, which may include a processor and a storage device, the storage device may be adapted to store a plurality of programs, and the programs may be adapted to be loaded by the processor to perform the target volume retrieval method described in the above-mentioned embodiment of the method.
It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements.
Furthermore, those skilled in the art will appreciate that while some embodiments described herein include some features included in other embodiments, rather than other features, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the claims of the present invention, any of the claimed embodiments may be used in any combination.
So far, the technical solutions of the present invention have been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of the present invention is obviously not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions can fall into the protection scope of the invention.
Claims (8)
1. A target volume retrieval method based on a target volume image is characterized by comprising the following steps:
acquiring image characteristics of a preset image training sample;
performing model training on a pre-constructed image retrieval model by using a machine learning algorithm based on a preset classification loss function and the image characteristics of the image training sample to obtain a preliminary optimization model;
performing model training on the preliminary optimization model by using a machine learning algorithm based on a preset double-threshold loss function and the image characteristics of the image training sample to obtain a final optimization model;
extracting image features of the target volume image based on the final optimization model;
calculating the similarity between the image characteristics of the target image and the image characteristics of the comparison images in a preset image library;
according to the calculation result, a comparison image similar to the target body image in the image library is obtained;
wherein the image retrieval model is based on a Simese network framework, and the dual-threshold loss function is represented by the following formula:
wherein, L (I)p,Iq) Represents a dual threshold loss function, said (I)p,Iq) Representing pairs of image samples, said alpha1Is the threshold value corresponding to the similar image sample pair, the alpha2A threshold corresponding to a dissimilar image sample pair; the y is a label value corresponding to the image sample pair, and y is 1 to indicate that the image sample pair is a similar image sample pair, and y is 0 to indicate that the image sample pair is a dissimilar image sample pair; d | | | f (I)p)-f(Iq)||2The | | | f (I)p)-f(Iq)||2Denotes f (I)p)-f(Iq) 2 norm of, said f (I)p) As an image IpThe image feature of f (I)q) As an image IqThe image feature of (1).
2. The target volume retrieval method based on the target volume image as claimed in claim 1, wherein before the step of calculating the similarity of the image features of the target volume image and the image features of the comparison images in the preset image library, the method further comprises:
performing dimension reduction processing on the image features of the target volume image according to the image feature dimensions of the compared images in the image library;
in this case, the step of "calculating the similarity between the image feature of the target volume image and the image feature of the comparison image in the preset image library" further includes calculating the similarity between the image feature of the target volume image after the dimension reduction processing and the image feature of the comparison image.
3. The target volume retrieval method based on the target volume image according to claim 2, wherein the step of performing dimension reduction processing on the image feature of the target volume image includes performing dimension reduction processing on the image feature of the target volume image by a principal component analysis method.
4. The target volume retrieval method based on the target volume image as claimed in any one of claims 1 to 3, wherein before the step of performing model training on the preliminary optimization model based on the preset dual-threshold loss function and the image features of the image training samples by using a machine learning algorithm to obtain a final optimization model, the method further comprises:
acquiring a similar image sample pair and a dissimilar image sample pair according to the image training sample;
respectively calculating the probability distribution of the characteristic distances of the similar image sample pairs and the probability distribution of the characteristic distances of the dissimilar image sample pairs;
and obtaining a threshold corresponding to the similar image sample pair and a threshold corresponding to the dissimilar image sample pair in the double-threshold loss function according to the probability distribution.
5. The target volume retrieval method based on the target volume image as claimed in any one of claims 1 to 3, wherein before the step of performing model training on the image retrieval model based on the preset classification loss function and the image features of the image training samples by using a machine learning algorithm to obtain a preliminary optimization model, the method further comprises:
acquiring the parameter weight of a neural network which finishes network training in advance, and performing parameter weight initialization on the image retrieval model according to the parameter weight;
the neural network is obtained by performing network training by using the machine learning algorithm based on the ImageNet data set.
6. The target volume image-based target volume retrieval method according to any one of claims 1 to 3, wherein the classification loss function is represented by the following formula:
Lc=-wklog(pk)
wherein, L iscRepresents a classification loss function, said wkWeight corresponding to the kth preset image category, pkThe probability that the image class of a certain image sample is the kth image class.
7. A storage device having stored therein a plurality of programs, characterized in that the programs are adapted to be loaded by a processor for performing the target volume image based target volume retrieval method according to any one of claims 1-6.
8. A control apparatus comprising a processor and a storage device, the storage device being adapted to store a plurality of programs,
characterized in that the program is adapted to be loaded by the processor to perform the target volume image based target volume retrieval method of any one of claims 1-6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810460265.8A CN108874889B (en) | 2018-05-15 | 2018-05-15 | Target body retrieval method, system and device based on target body image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810460265.8A CN108874889B (en) | 2018-05-15 | 2018-05-15 | Target body retrieval method, system and device based on target body image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108874889A CN108874889A (en) | 2018-11-23 |
CN108874889B true CN108874889B (en) | 2021-01-12 |
Family
ID=64334424
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810460265.8A Active CN108874889B (en) | 2018-05-15 | 2018-05-15 | Target body retrieval method, system and device based on target body image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108874889B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919208A (en) * | 2019-02-25 | 2019-06-21 | 中电海康集团有限公司 | A kind of appearance images similarity comparison method and system |
CN109949229A (en) * | 2019-03-01 | 2019-06-28 | 北京航空航天大学 | A kind of target cooperative detection method under multi-platform multi-angle of view |
CN110147461A (en) * | 2019-04-30 | 2019-08-20 | 维沃移动通信有限公司 | Image display method, device, terminal device and computer readable storage medium |
CN110674855B (en) * | 2019-09-12 | 2021-03-09 | 创新奇智(重庆)科技有限公司 | Classification loss function formulation method based on flaw area of light-coated part |
EP3872695A1 (en) * | 2020-02-28 | 2021-09-01 | Toyota Jidosha Kabushiki Kaisha | A method and system of adapting an initial model of a neural network |
CN113760415A (en) * | 2020-06-04 | 2021-12-07 | Oppo广东移动通信有限公司 | Dial plate generation method and device, electronic equipment and computer readable storage medium |
CN113139653B (en) * | 2021-03-18 | 2024-06-14 | 有米科技股份有限公司 | Neural network training method and device for image hash solving |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105825163A (en) * | 2015-01-09 | 2016-08-03 | 杭州海康威视数字技术股份有限公司 | Retrieval system and method of face image |
US9792492B2 (en) * | 2015-07-07 | 2017-10-17 | Xerox Corporation | Extracting gradient features from neural networks |
CN106407281B (en) * | 2016-08-26 | 2019-12-24 | 北京奇艺世纪科技有限公司 | Image retrieval method and device |
CN106649665A (en) * | 2016-12-14 | 2017-05-10 | 大连理工大学 | Object-level depth feature aggregation method for image retrieval |
-
2018
- 2018-05-15 CN CN201810460265.8A patent/CN108874889B/en active Active
Non-Patent Citations (3)
Title |
---|
Dual threshold based unsupervised face image clustering;qiang Deng等;《IEEE》;20101231;正文第1-5页 * |
Image edge detection algorithm based on the parallel genetic and otsu dual-threshold;Dang X等;《International conference on intelligent systems research & mechatronics engineering 2015》;20151231;正文第1-4页 * |
特殊靶标图像的Otsu自适应双阈值分割法;王召利等;《知网》;20101231;正文第1-3页 * |
Also Published As
Publication number | Publication date |
---|---|
CN108874889A (en) | 2018-11-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108874889B (en) | Target body retrieval method, system and device based on target body image | |
CN110781276B (en) | Text extraction method, device, equipment and storage medium | |
CN111860573B (en) | Model training method, image category detection method and device and electronic equipment | |
CN106683680B (en) | Speaker recognition method and device, computer equipment and computer readable medium | |
US10796244B2 (en) | Method and apparatus for labeling training samples | |
CN109978060B (en) | Training method and device of natural language element extraction model | |
CN112559800B (en) | Method, apparatus, electronic device, medium and product for processing video | |
CN110489521B (en) | Text type detection method and device, electronic equipment and computer readable medium | |
CN113254654B (en) | Model training method, text recognition method, device, equipment and medium | |
JP5523543B2 (en) | Concept recognition method and concept recognition device based on co-learning | |
CN110135505B (en) | Image classification method and device, computer equipment and computer readable storage medium | |
CN113065525B (en) | Age identification model training method, face age identification method and related device | |
CN110704597B (en) | Dialogue system reliability verification method, model generation method and device | |
CN110263218B (en) | Video description text generation method, device, equipment and medium | |
CN114241505B (en) | Method and device for extracting chemical structure image, storage medium and electronic equipment | |
CN109783805B (en) | Network community user identification method and device and readable storage medium | |
CN108959474A (en) | Entity relationship extracting method | |
CN112183102A (en) | Named entity identification method based on attention mechanism and graph attention network | |
CN114639096B (en) | Text recognition method, device, electronic equipment and storage medium | |
CN114429635A (en) | Book management method | |
CN110197213B (en) | Image matching method, device and equipment based on neural network | |
CN116935411A (en) | Radical-level ancient character recognition method based on character decomposition and reconstruction | |
CN112989040B (en) | Dialogue text labeling method and device, electronic equipment and storage medium | |
CN115222047A (en) | Model training method, device, equipment and storage medium | |
JP2017538226A (en) | Scalable web data extraction |
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 |