CN113033587B - Image recognition result evaluation method and device, electronic equipment and storage medium - Google Patents
Image recognition result evaluation method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN113033587B CN113033587B CN201911347791.4A CN201911347791A CN113033587B CN 113033587 B CN113033587 B CN 113033587B CN 201911347791 A CN201911347791 A CN 201911347791A CN 113033587 B CN113033587 B CN 113033587B
- Authority
- CN
- China
- Prior art keywords
- image
- identified
- features
- similarity
- result
- 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
- 238000011156 evaluation Methods 0.000 title claims abstract description 153
- 238000004364 calculation method Methods 0.000 claims abstract description 91
- 238000010606 normalization Methods 0.000 claims abstract description 29
- 238000000034 method Methods 0.000 claims abstract description 15
- 238000013507 mapping Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 11
- 230000000694 effects Effects 0.000 abstract description 6
- 238000012545 processing Methods 0.000 abstract description 6
- 238000004422 calculation algorithm Methods 0.000 abstract description 3
- 239000013598 vector Substances 0.000 description 28
- 238000000605 extraction Methods 0.000 description 13
- 239000011159 matrix material Substances 0.000 description 11
- 230000006870 function Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 238000001514 detection method Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000013209 evaluation strategy Methods 0.000 description 3
- 238000001303 quality assessment method Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000001815 facial effect Effects 0.000 description 2
- 239000011521 glass Substances 0.000 description 2
- 238000013441 quality evaluation Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention provides an image recognition result evaluation method, an image recognition result evaluation device, electronic equipment and a storage medium, wherein the method comprises the following steps: acquiring image features to be identified and evaluation factors of the image features to be identified; carrying out normalization calculation on the image features to be identified to obtain normalization features; fusing the normalized features with the evaluation factors to obtain similarity calculation features; performing similarity calculation on the similarity calculation features and the bottom library image features in the database to obtain a similarity result, and processing the bottom library image features and the similarity calculation features through the same algorithm and storing the processed bottom library image features and the similarity calculation features in the database; and evaluating the recognition result of the image feature to be recognized based on the similarity result. By adding the evaluation factors to the normalized image features to be identified, the features combined with the evaluation scale have better identification effect than the commonly used normalized features.
Description
Technical Field
The present invention relates to the field of data processing technologies, and in particular, to an image recognition result evaluation method, an image recognition result evaluation device, an electronic device, and a storage medium.
Background
With the deep research of image and image recognition technology, more and more image recognition products are developed and applied, such as face recognition, pedestrian re-recognition and the like. In the image recognition process, an image is input into a feature extraction network, the image features of the image are extracted, the image is classified and recognized through the image features, and the output is a classification result. However, because the images are affected by illumination, shooting angles, image shielding, acquisition equipment differences and other factors, the quality differences of the images acquired by the actual application scenes are very large, but common normalized features do not have the capability of representing the image quality, and the features extracted from the images with different qualities are judged by using common judgment standards, so that the credibility of the identification results can be affected.
Disclosure of Invention
The embodiment of the invention provides an image recognition result evaluation method which can improve the evaluation effect of a recognition result.
In a first aspect, an embodiment of the present invention provides an image recognition result evaluation method, including:
Acquiring image features to be identified and evaluation factors of the image features to be identified;
carrying out normalization calculation on the image features to be identified to obtain normalization features;
Fusing the normalized features with the evaluation factors to obtain similarity calculation features;
Performing similarity calculation on the similarity calculation features and the bottom library image features in the database to obtain a similarity result, wherein the bottom library image features and the images to be identified are obtained through the same calculation;
and evaluating the recognition result of the image feature to be recognized based on the similarity result.
Optionally, the acquiring the evaluation factor of the image feature to be identified includes:
and carrying out dot product operation and mapping on the image features to be identified and the weight factors trained in advance to obtain evaluation factors of the image features to be identified, wherein the evaluation factors of the image features to be identified are mapped into values of preset intervals.
Optionally, the evaluation factor is a quality confidence, the pre-trained weight factor is a pre-trained quality weight, the dot product operation is performed on the image feature to be identified and the pre-trained weight factor to obtain the evaluation factor of the image feature to be identified, and the evaluation factor of the image feature to be identified is mapped to a value of a preset interval, including:
and carrying out dot product operation and mapping on the image features to be identified and the pre-trained quality weights to obtain the quality confidence coefficient of the image features to be identified.
Optionally, the normalizing calculation is performed on the image feature to be identified to obtain a normalized feature, which includes:
extracting each characteristic value in the image characteristics to be identified, and calculating the square sum of each characteristic value;
calculating to obtain a square root result of the square sum;
And normalizing the image features to be identified based on a square root result to obtain normalized features of the image features to be identified.
Optionally, the evaluating the recognition result of the image feature to be recognized based on the similarity result includes:
Comparing the similarity result with a preset similarity threshold value, and judging whether the similarity result is larger than the preset similarity threshold value or not;
if the similarity result is larger than the preset similarity result, evaluating the similarity result as a desired result;
And if the similarity result is smaller than the preset similarity result, evaluating that the similarity result is not a desired result.
Optionally, before comparing the similarity result with a preset similarity threshold, the method further includes:
Acquiring item information corresponding to the image features to be identified;
And matching a preset similarity threshold according to the item information.
In a second aspect, an embodiment of the present invention provides an image recognition result evaluation apparatus, including:
The first acquisition module is used for acquiring the image characteristics to be identified and the evaluation factors of the image characteristics to be identified;
The first calculation module is used for carrying out normalization calculation on the image features to be identified to obtain normalization features;
The second calculation module is used for fusing the normalized features with the evaluation factors to obtain similarity calculation features;
The third calculation module is used for carrying out similarity calculation on the similarity calculation characteristics and the bottom library image characteristics in the database to obtain a similarity result, wherein the bottom library image characteristics and the image to be identified are obtained through the same calculation;
and the evaluation module is used for evaluating the recognition result of the image feature to be recognized based on the similarity result.
In a third aspect, an embodiment of the present invention provides an electronic device, including: the image recognition result evaluation method comprises the steps of a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein the steps in the image recognition result evaluation method provided by the embodiment of the invention are realized when the processor executes the computer program.
In a fourth aspect, an embodiment of the present invention provides a computer readable storage medium, where a computer program is stored, where the computer program when executed by a processor implements the steps in the image recognition result evaluation method provided by the embodiment of the present invention.
In the embodiment of the invention, the image characteristics to be identified and the evaluation factors of the image characteristics to be identified are obtained; carrying out normalization calculation on the image features to be identified to obtain normalization features; fusing the normalized features with the evaluation factors to obtain similarity calculation features; performing similarity calculation on the similarity calculation features and the bottom library image features in the database to obtain a similarity result, and processing the bottom library image features and the similarity calculation features through the same algorithm and storing the processed bottom library image features and the similarity calculation features in the database; and evaluating the recognition result of the image feature to be recognized based on the similarity result. By adding the evaluation factors to the normalized image features to be identified, the image features to be identified have an evaluation scale, and the features combined with the evaluation scale have better identification effect than the commonly used normalized features.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of an image recognition result evaluation method according to an embodiment of the present invention;
FIG. 2 is a flowchart of another image recognition result evaluation method according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of an image recognition result evaluation device according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of another image recognition result evaluation device according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of another image recognition result evaluation device according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of another image recognition result evaluation device according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, fig. 1 is a flowchart of an image recognition result evaluation method according to an embodiment of the present invention, as shown in fig. 1, including the following steps:
101. and acquiring the image characteristics to be identified and the evaluation factors of the image characteristics to be identified.
The image features to be identified may be image features obtained by collecting a target image by an image collecting device (such as a camera) and extracting features of the target image, such as face image features, vehicle image features or other object image features; the image features to be identified can also be image features obtained by uploading the target image to an application through an uploading window and extracting the features of the target image; the image feature to be identified may also be an image feature obtained by extracting a feature of a target image obtained by selecting the target image in the image set by a user. In some possible embodiments, the image features can also be directly uploaded or selected by the user.
The evaluation factor of the image feature to be identified is used for evaluating the credibility of the image feature to be identified, the evaluation factor can be obtained through the image feature to be identified, specifically, the image feature to be identified and the weight factor trained in advance are subjected to dot product operation, the result after the dot product operation is mapped to a preset interval, for example, the result after the dot product operation can be mapped to the (0, 1) interval through a function with the value range of (0, 1) interval, so as to obtain the corresponding evaluation factor, and the function with the value range of (0, 1) interval can be a logic function, for example, a Sigmoid function. It should be noted that the image feature to be identified may be in a matrix form or a vector form, the weight factor may be in a matrix form or a vector form, and the evaluation factor may be in a scalar form. For example, when the feature of the image to be identified is 1*n, the pre-trained weight factor is n×1 weight vector, and the dot product operation and Sigmoid function mapping are performed on the 1*n feature vector and n×1 weight vector, so as to obtain a scalar of (0, 1) interval, i.e. the evaluation factor. When the feature of the image to be identified is m×n feature matrix, the m×n feature matrix can be understood as being composed of m 1*n feature vectors, the pre-trained weight factor is n×1 weight vector, the dot product operation is performed on the m×n feature vector and the n×1 weight vector, so as to obtain m×1 scalar, the scalar is mapped to the (0, 1) interval through the Sigmoid function to obtain new scalar, and each mapped new scalar corresponds to one 1*n feature vector, namely the evaluation factor.
The evaluation factor of the image feature to be identified may be a quality confidence coefficient of the image to be identified, where the quality confidence coefficient is used to represent a confidence level of the image to be identified in an image quality dimension, and the higher the quality confidence coefficient is, the higher the image quality is, and the higher the accuracy of the result of identifying using the image to be identified is. Correspondingly, the weight factors are pre-trained quality weights. The quality confidence may be a single quality confidence of a single attribute, such as image light, image occlusion, image size, or image pixel quality, or a comprehensive quality confidence of at least two of image light, image occlusion, image size, and image pixel quality. For taking the image quality confidence coefficient of a single attribute as an evaluation factor, the evaluation factor can be obtained by extracting through an evaluation feature extraction model, for example, the image light information in the image to be identified can be extracted through an image light feature extraction model, so as to obtain corresponding image light as the evaluation factor; extracting image shielding information in an image to be identified through an object detection model and calculating shielding rate to serve as an evaluation factor; calculating the image size information of the image to be identified through an image size calculation model to obtain the image size of the image to be identified as an evaluation factor; the pixel quality information of the image to be identified can be extracted through the image pixel quality evaluation model, and the pixel quality of the image to be identified is obtained to serve as an evaluation factor. For taking the comprehensive image quality confidence coefficient as an evaluation factor, the evaluation factors can be respectively extracted by the method, and respectively weighted and summed after extraction to obtain corresponding evaluation factors, for example, the image light is A, the image shielding is B, the image size is C, the image pixel quality is D, and the corresponding weighting coefficients are 0.2, 0.3, 0.1 and 0.4 respectively, then the evaluation factor Z is 0.2A+0.3B+0.1C+0.4D, wherein the sum of A, B, C and D is equal to 1, and the sum is equivalent to the value of A, B, C and D after normalization of the image quality confidence coefficient of each single attribute.
In the embodiment of the invention, the comprehensive image quality confidence coefficient may be used as an evaluation factor, or the quality confidence coefficient of the image to be identified may be extracted according to a comprehensive quality evaluation model, which may be pre-trained. In the training process of the comprehensive quality assessment model, each image in the training data set can be marked with a scoring label, the scoring label is a floating point number label of 0-1, and the training data set marked with the scoring label is input into the comprehensive quality assessment model for training, so that the comprehensive quality assessment learns the score of the image. After the image features to be identified are input, in the model, dot product operation is carried out on the image features to be identified and the trained quality weights, and the dot product operation is mapped to a (0, 1) interval through a Sigmoid function, so that the quality confidence coefficient of the image features to be identified is obtained.
102. And carrying out normalization calculation on the image features to be identified to obtain normalization features.
In this step, the normalization of the image feature to be identified may be understood as mapping the values in the image feature to be identified to a range of values of [0,1] or [ -1,1] by a normalization function, so that the dimensional expression becomes a non-dimensional expression, for example, mapping the pixel values with dimensions to floating point numbers within the range of [0,1] or [ -1,1] without dimensions.
The normalization of the image features to be identified can be based on the sum of absolute values, the normalization calculation can be based on the absolute value of the error, the sum of all feature values in the image features to be identified can be calculated, and then the ratio of each feature of the image features to be identified to the sum of all feature values is calculated, and the ratio is used as the feature value in the normalized feature.
The normalization of the features of the image to be identified may be performed based on a square sum, and in the embodiment of the present invention, normalization calculation is preferably performed on the image to be identified based on the square of the error. Specifically, each feature value in the image features to be identified can be extracted, and the square sum of each feature value is calculated; then calculating to obtain square root result of the square sum; normalizing the image features to be identified based on the square root result to obtain normalized features of the image features to be identified. After the square root result is calculated, calculating the ratio of each eigenvalue of the image feature to be identified to the square root, and taking the ratio as the eigenvalue in the normalized feature, wherein the specific formula is as follows:
The f norm is the normalized feature of the image feature to be identified, the f is the image feature to be identified, and the n is the number of feature values in the image feature to be identified. And normalizing the image to be identified based on the square sum, and avoiding the occurrence of the over-fitting phenomenon due to smaller normalized parameter values.
103. And fusing the normalized features with the evaluation factors to obtain similarity calculation features.
In this step, the normalized feature may be a feature in a matrix form or a vector form, the evaluation factor is a scalar, and the obtained similarity calculation feature is also a feature in a matrix form or a vector form by fusing the normalized feature with the evaluation factor. The fusion may be multiplicative or exponential, and specifically, the normalized feature may be multiplied or added to the evaluation factor, or the normalized feature may be fused by taking the evaluation factor as an index of the value in the normalized feature.
When the normalized feature is a vector, the normalized feature is multiplied by an evaluation factor, and the obtained similarity calculation feature is a feature in the form of a vector, for example, assuming that the evaluation factor is a and the normalized feature is a vector (0.8,0.6), the similarity calculation feature is a vector of a× (0.8,0.6) = (0.8 a,0.6 a). When the evaluation factor is an index of values in the normalized features, the similarity calculation feature is a vector of (0.8 a,0.6a).
When the normalized feature is a matrix, the normalized feature is multiplied by an evaluation factor, and the obtained similarity calculation feature is a feature in a matrix form. For example, assuming that the normalized feature is a feature matrix of m×n, the feature matrix is composed of m feature vectors, the evaluation factors are m×1, the m evaluation factors are multiplied by the m feature vectors respectively, further assuming that the normalized feature is composed of a first feature vector (0.8,0.6) and a second feature vector (0.6,0.8), the evaluation factors are a first evaluation factor a, a second evaluation factor b, and the first feature vector and the second feature vector are respectively corresponding to each other, the evaluation factors are multiplied by the feature vectors respectively to obtain the following similarity calculation feature:
it should be noted that, the above similarity calculation feature is used for representing the image to be identified and the features of the bottom library image to calculate, and the similarity calculation feature can have an evaluation scale due to the addition of the evaluation factor to the similarity calculation feature.
104. And carrying out similarity calculation on the similarity calculation features and the image features of the database midsole database to obtain a similarity result.
In this step, the similarity calculation feature is calculated in step 103, and the image feature of the database midsole library and the image to be identified are obtained by the same normalization calculation.
The bottom library image features refer to image features stored in a database, and the image features are used for carrying out similarity calculation with the image features to be identified, so that which image feature to be identified is most similar to the image features according to the calculated similarity, and further information of the image to be identified is obtained according to the most similar image features. The database stores structured data or semi-structured data, where the structured data and the semi-structured data include a base image feature and other information corresponding to the base image feature, for example, when the base image feature is a face feature, the other information may include identity information, contact information, and the like, and when the base image feature is vehicle information, the other information may include vehicle brand information, vehicle model information, and the like.
The bottom library image features may be normalized image features obtained by capturing a target image by an image capturing device (such as a camera) and extracting features of the target image, such as normalized face image features, normalized vehicle image features, or normalized image features of other objects. Specifically, after the target image is acquired through the image acquisition device, the feature extraction engine is used for extracting the features of the target image to obtain corresponding feature vectors, and the feature vectors are normalized and then stored in the database. Of course, when there are multiple target images, after feature extraction of the target images is performed by the feature extraction engine, multiple feature vectors are obtained, for example, when the target images include faces, glasses and caps, the face image features, the glasses image features and the vectors corresponding to the cap image features are extracted respectively to form a matrix form of features.
In a possible embodiment, the above-mentioned bottom-library image features also have an evaluation factor, and the calculation method of the evaluation factor of the bottom-library image features in step 101 is not described herein. It should be noted that, in the case where the image feature of the base has the evaluation factor, the calculation method of the evaluation factor of the image feature to be identified needs to be the same as the calculation method of the evaluation factor of the image feature of the base. Since the similarity calculation features and the image features in the bottom library have evaluation factors, namely the similarity calculation features and the image features in the bottom library have evaluation scales, the evaluation scales are more obvious after the similarity calculation.
In another possible embodiment, the above-mentioned evaluation factor of the bottom library image feature needs to be greater than a preset storage threshold, and after the calculation of the evaluation factor of the target image feature, the target image feature is stored as the bottom library image feature only when the evaluation factor is greater than the preset storage threshold, so as to ensure the quality of the bottom library image feature with the evaluation scale.
The similarity uses the inner product of the features, which is equivalent to the cosine similarity calculated by two normalized features and then multiplied by two feature evaluation factors.
105. And evaluating the recognition result of the image features to be recognized based on the similarity result.
In this step, the similarity result is obtained by calculating the similarity calculation feature and the bottom library image feature, so that an evaluation scale of the dimension corresponding to the evaluation factor can be directly expressed, the abstract reliability is quantified, the similarity result can be evaluated according to the evaluation scale, and whether the similarity result is a desired similarity result is evaluated.
The above-mentioned recognition results are characterized by similarity results, for example, the similarity result corresponding to the maximum similarity may be taken as the recognition result of the image feature to be recognized, or the similarity result with the similarity greater than a certain set value may be taken as the recognition result of the image feature to be recognized.
Alternatively, the similarity result may be evaluated by a preset similarity threshold. Comparing the similarity result with a preset similarity threshold value, and judging whether the similarity result is larger than the preset similarity threshold value or not; if the similarity result is larger than the preset similarity result, evaluating the similarity result as a desired result; if the similarity result is smaller than the preset similarity result, the similarity result is not the expected result. For example, the evaluation factor is a quality confidence, where the quality confidence is used to characterize the image quality of the image to be recognized, and may be the image quality of a single attribute or the image quality of a comprehensive attribute, and when the image quality of the image to be recognized is low, the reliability of the recognition result characterized by the calculated similarity result is low, and may be regarded as a non-desired result, and when the image quality of the image to be recognized is high, the reliability of the recognition result characterized by the calculated similarity result is high, and may be regarded as a desired result.
In the embodiment of the invention, the image characteristics to be identified and the evaluation factors of the image characteristics to be identified are obtained; carrying out normalization calculation on the image features to be identified to obtain normalization features; fusing the normalized features with the evaluation factors to obtain similarity calculation features; performing similarity calculation on the similarity calculation features and the bottom library image features in the database to obtain a similarity result, and processing the bottom library image features and the similarity calculation features through the same algorithm and storing the processed bottom library image features and the similarity calculation features in the database; and evaluating the recognition result of the image feature to be recognized based on the similarity result. By adding the evaluation factors to the normalized image features to be identified, the image features to be identified have an evaluation scale, and the features combined with the evaluation scale have better identification effect than the commonly used normalized features.
It should be noted that, the image recognition result evaluation method provided by the embodiment of the invention can be applied to devices such as a mobile phone, a monitor, a computer, a server and the like which need to evaluate the image recognition result.
Specifically, referring to fig. 2, fig. 2 is a flowchart of an image recognition result evaluation method provided by an embodiment of the present invention, as shown in fig. 2, including:
201. and acquiring the image characteristics to be identified.
202. And carrying out dot product operation and mapping on the image features to be identified and the evaluation factor weights to obtain the evaluation factors of the image features to be identified.
203. And carrying out normalization calculation on the image features to be identified to obtain normalization features.
204. And multiplying the normalized feature by an evaluation factor to obtain a similarity calculation feature.
205. And carrying out similarity calculation on the similarity calculation features and the image features of the database midsole database to obtain a similarity result.
206. And acquiring item information corresponding to the image features to be identified.
In this step, the item information is an image recognition item, and the image recognition item may be specific application item information for image recognition such as a face recognition item, a pedestrian re-recognition item, an object detection item, an object classification item, and a vehicle detection item.
The item information may be obtained by selecting by a user, and because parameters in feature extraction engines corresponding to different item types are different, the item information may also be determined by the type of the feature extraction engine that performs feature extraction on the image to be identified, for example, a face image in a face recognition item is extracted by the face feature extraction engine, and a vehicle image in a vehicle detection item is extracted by the vehicle feature extraction engine.
In one possible embodiment, different databases may be built for different image recognition items, corresponding base image features are stored, and the item information may also be obtained through the selected databases. For example, if the database stores facial image features, the item information may be determined to be a facial recognition item.
207. And matching a preset similarity threshold according to the item information.
In this step, because the image quality required by different projects is different, the fault tolerance requirements of different projects on the recognition results are also different, so that different evaluations can be performed on the recognition results of the image features according to different project information, that is, different evaluation strategies are set to evaluate whether the image to be recognized meets the requirements of the corresponding image recognition projects. The above-described evaluation strategy may be a similarity threshold for comparison with a similarity result.
In a possible embodiment, the preset evaluation policy may be pre-stored in a corresponding database and associated with the database, and when the database is selected, the corresponding evaluation policy is correspondingly matched.
208. And comparing the similarity result with a preset similarity threshold value, and judging whether the similarity result is larger than the preset similarity threshold value or not.
209. And if the similarity result is larger than the preset similarity result, evaluating the similarity result as a desired result.
210. If the similarity result is smaller than the preset similarity result, the similarity result is not the expected result.
In the steps 208, 209, 210, after calculating the similarity result, it may be evaluated whether the recognition result of the image feature to be recognized meets the recognition result expected by the corresponding item, so as to perform subsequent processing on the recognition result, where the subsequent processing includes discarding, outputting the evaluation result, outputting the recognition result, and so on. For example, the recognition result which does not meet the requirement of the corresponding item is discarded, and the recognition result which meets the requirement of the item is output.
In the embodiment of the invention, different evaluation strategies are set according to different projects, so that the evaluation result is more accurate and is closer to the project requirements, and the evaluation effect for different projects is further improved.
It should be noted that, the image recognition result evaluation method provided by the embodiment of the invention can be applied to devices such as a mobile phone, a monitor, a computer, a server and the like which need to evaluate the image recognition result.
Referring to fig. 3, fig. 3 is a schematic structural diagram of an image recognition result evaluation device according to an embodiment of the present invention, as shown in fig. 3, the device includes:
a first obtaining module 301, configured to obtain an image feature to be identified and an evaluation factor of the image feature to be identified;
the first calculation module 302 is configured to perform normalization calculation on the image feature to be identified, so as to obtain a normalized feature;
A second calculation module 303, configured to fuse the normalized feature with the evaluation factor to obtain a similarity calculation feature;
The third calculation module 304 is configured to perform similarity calculation on the similarity calculation feature and a bottom library image feature in the database, so as to obtain a similarity result, where the bottom library image feature and the image to be identified are obtained by the same calculation;
And the evaluation module 305 is configured to evaluate the recognition result of the image feature to be recognized based on the similarity result.
Optionally, the first obtaining module 301 is further configured to perform dot product operation and mapping on the image feature to be identified and a weight factor trained in advance, so as to obtain an evaluation factor of the image feature to be identified, where the evaluation factor of the image feature to be identified is mapped to a value of a preset interval.
Optionally, the evaluation factor is a quality confidence coefficient, the pre-trained weight factor is a pre-trained quality weight, and the first obtaining module 301 is further configured to perform dot product operation and mapping on the image feature to be identified and the pre-trained quality weight to obtain the quality confidence coefficient of the image feature to be identified, where the quality confidence coefficient of the image feature to be identified is a value of a (0, 1) interval.
Optionally, as shown in fig. 4, the first computing module 302 includes:
an extraction unit 3021 for extracting each feature value in the image feature to be identified, and calculating a sum of squares of the each feature value;
a first calculation unit 3022 for calculating a square root result of the sum of squares;
And the second calculating unit 3023 is configured to normalize the image feature to be identified based on the square root result, and obtain a normalized feature of the image feature to be identified.
Optionally, as shown in fig. 5, the evaluation module 305 includes:
A comparing unit 3051, configured to compare the similarity result with a preset similarity threshold, and determine whether the similarity result is greater than the preset similarity threshold;
a first evaluation unit 3052, configured to evaluate the similarity result as a desired result if the similarity result is greater than the preset similarity result;
And the second evaluation unit 3053 is configured to evaluate that the similarity result is not the desired result if the similarity result is smaller than the preset similarity result.
Optionally, as shown in fig. 6, the apparatus further includes:
A second obtaining module 306, configured to obtain item information corresponding to the image feature to be identified;
And a matching module 307, configured to match a preset similarity threshold according to the item information.
It should be noted that, the image recognition result evaluation device provided by the embodiment of the invention can be applied to devices such as a mobile phone, a monitor, a computer, a server and the like which need to evaluate the image recognition result.
The image recognition result evaluation device provided by the embodiment of the invention can realize each process realized by the image recognition result evaluation method in the method embodiment, and can achieve the same beneficial effects. In order to avoid repetition, a description thereof is omitted.
Referring to fig. 7, fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, as shown in fig. 7, including: a memory 702, a processor 701 and a computer program stored on the memory 702 and executable on the processor 701, wherein:
the processor 701 is configured to call a computer program stored in the memory 702, and perform the following steps:
Acquiring image features to be identified and evaluation factors of the image features to be identified;
carrying out normalization calculation on the image features to be identified to obtain normalization features;
Fusing the normalized features with the evaluation factors to obtain similarity calculation features;
Performing similarity calculation on the similarity calculation features and the bottom library image features in the database to obtain a similarity result, wherein the bottom library image features and the images to be identified are obtained through the same calculation;
and evaluating the recognition result of the image feature to be recognized based on the similarity result.
Optionally, the acquiring the evaluation factor of the image feature to be identified performed by the processor 701 includes:
And carrying out dot product operation and mapping on the image features to be identified and the weight factors trained in advance to obtain evaluation factors of the image features to be identified, wherein the evaluation factors of the image features to be identified are mapped into values of (0, 1) intervals.
Optionally, the evaluation factor is a quality confidence, the pre-trained weight factor is a pre-trained quality weight, the performing, by the processor 701, dot product operation and mapping the image feature to be identified and the pre-trained weight factor to obtain the evaluation factor of the image feature to be identified, where the performing includes:
And carrying out dot product operation and mapping on the image features to be identified and the pre-trained quality weights to obtain the quality confidence coefficient of the image features to be identified, wherein the quality confidence coefficient of the image features to be identified is a value of a (0, 1) interval.
Optionally, the normalizing calculation performed by the processor 701 on the image feature to be identified obtains a normalized feature, including:
extracting each characteristic value in the image characteristics to be identified, and calculating the square sum of each characteristic value;
calculating to obtain a square root result of the square sum;
And normalizing the image features to be identified based on a square root result to obtain normalized features of the image features to be identified.
Optionally, the evaluating, by the processor 701, the recognition result of the image feature to be recognized based on the similarity result includes:
Comparing the similarity result with a preset similarity threshold value, and judging whether the similarity result is larger than the preset similarity threshold value or not;
if the similarity result is larger than the preset similarity result, evaluating the similarity result as a desired result;
And if the similarity result is smaller than the preset similarity result, evaluating that the similarity result is not a desired result.
Optionally, before the processor 701 performs the comparing the similarity result with a preset similarity threshold, the processor 701 performs further includes:
Acquiring item information corresponding to the image features to be identified;
And matching a preset similarity threshold according to the item information.
The electronic device may be a mobile phone, a monitor, a computer, a server, or the like, which is used to evaluate the image recognition result.
The electronic device provided by the embodiment of the invention can realize each process realized by the image recognition result evaluation method in the embodiment of the method, can achieve the same beneficial effects, and is not repeated here for avoiding repetition.
The embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements each process of the image recognition result evaluation method provided by the embodiment of the invention, and can achieve the same technical effect, so that repetition is avoided, and no further description is provided herein.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM) or the like.
The foregoing disclosure is illustrative of the present invention and is not to be construed as limiting the scope of the invention, which is defined by the appended claims.
Claims (5)
1. An image recognition result evaluation method is characterized by comprising the following steps:
Acquiring image features to be identified and evaluation factors of the image features to be identified;
carrying out normalization calculation on the image features to be identified to obtain normalization features;
Fusing the normalized features with the evaluation factors to obtain similarity calculation features;
Performing similarity calculation on the similarity calculation features and the bottom library image features in the database to obtain a similarity result, wherein the bottom library image features and the images to be identified are obtained through the same calculation;
evaluating the recognition result of the image feature to be recognized based on the similarity result;
The acquiring the evaluation factor of the image feature to be identified comprises the following steps:
Performing dot product operation and mapping on the image features to be identified and the weight factors trained in advance to obtain evaluation factors of the image features to be identified, wherein the evaluation factors of the image features to be identified are mapped into values of preset intervals;
The evaluation factor is a quality confidence, the pre-trained weight factor is a pre-trained quality weight, the dot product operation and mapping are performed on the image feature to be identified and the pre-trained weight factor to obtain the evaluation factor of the image feature to be identified, and the method comprises the following steps:
Performing dot product operation and mapping on the image features to be identified and the pre-trained quality weights to obtain the quality confidence coefficient of the image features to be identified;
The evaluating the recognition result of the image feature to be recognized based on the similarity result comprises the following steps:
Comparing the similarity result with a preset similarity threshold value, and judging whether the similarity result is larger than the preset similarity threshold value or not;
if the similarity result is larger than the preset similarity result, evaluating the similarity result as a desired result;
And if the similarity result is smaller than the preset similarity result, evaluating that the similarity result is not a desired result.
2. The method of claim 1, wherein prior to said comparing the similarity result to a pre-set similarity threshold, the method further comprises:
Acquiring item information corresponding to the image features to be identified;
And matching a preset similarity threshold according to the item information.
3. An image recognition result evaluation apparatus, characterized by comprising:
The first acquisition module is used for acquiring the image characteristics to be identified and the evaluation factors of the image characteristics to be identified;
The first calculation module is used for carrying out normalization calculation on the image features to be identified to obtain normalization features;
The second calculation module is used for fusing the normalized features with the evaluation factors to obtain similarity calculation features;
The third calculation module is used for carrying out similarity calculation on the similarity calculation characteristics and the bottom library image characteristics in the database to obtain a similarity result, wherein the bottom library image characteristics and the image to be identified are obtained through the same calculation;
the evaluation module is used for evaluating the recognition result of the image feature to be recognized based on the similarity result;
The first acquisition module is further used for carrying out dot product operation and mapping on the image features to be identified and the weight factors trained in advance so as to obtain evaluation factors of the image features to be identified, and the evaluation factors of the image features to be identified are mapped into values of preset intervals;
the evaluation factor is quality confidence, the pre-trained weight factor is pre-trained quality weight, and the first acquisition module is further configured to perform dot product operation and mapping on the image feature to be identified and the pre-trained quality weight to obtain quality confidence of the image feature to be identified;
The evaluation module comprises:
the comparison unit is used for comparing the similarity result with a preset similarity threshold value and judging whether the similarity result is larger than the preset similarity threshold value or not;
the first evaluation unit is used for evaluating the similarity result to be a desired result if the similarity result is larger than the preset similarity result;
And the second evaluation unit is used for evaluating that the similarity result is not a desired result if the similarity result is smaller than the preset similarity result.
4. An electronic device, comprising: memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in the image recognition result evaluation method according to claim 1 or 2 when the computer program is executed.
5. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when executed by a processor, implements the steps of the image recognition result evaluation method according to claim 1 or 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911347791.4A CN113033587B (en) | 2019-12-24 | 2019-12-24 | Image recognition result evaluation method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911347791.4A CN113033587B (en) | 2019-12-24 | 2019-12-24 | Image recognition result evaluation method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113033587A CN113033587A (en) | 2021-06-25 |
| CN113033587B true CN113033587B (en) | 2024-06-11 |
Family
ID=76451666
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911347791.4A Active CN113033587B (en) | 2019-12-24 | 2019-12-24 | Image recognition result evaluation method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113033587B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113592818A (en) * | 2021-07-30 | 2021-11-02 | 北京小米移动软件有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN117315301B (en) * | 2023-10-07 | 2024-04-09 | 长春理工大学 | Collection information matching system and method based on image recognition |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109214273A (en) * | 2018-07-18 | 2019-01-15 | 平安科技（深圳）有限公司 | Facial image comparison method, device, computer equipment and storage medium |
| WO2019100608A1 (en) * | 2017-11-21 | 2019-05-31 | 平安科技（深圳）有限公司 | Video capturing device, face recognition method, system, and computer-readable storage medium |
-
2019
- 2019-12-24 CN CN201911347791.4A patent/CN113033587B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019100608A1 (en) * | 2017-11-21 | 2019-05-31 | 平安科技（深圳）有限公司 | Video capturing device, face recognition method, system, and computer-readable storage medium |
| CN109214273A (en) * | 2018-07-18 | 2019-01-15 | 平安科技（深圳）有限公司 | Facial image comparison method, device, computer equipment and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 基于纹理特征融合的人脸图像质量评估算法;陈正浩;吴云东;蔡国榕;陈水利;;集美大学学报(自然科学版)(04);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113033587A (en) | 2021-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112990432B (en) | Target recognition model training method and device and electronic equipment | |
| CN108108807B (en) | Learning type image processing method, system and server | |
| CN112348117B (en) | Scene recognition method, device, computer equipment and storage medium | |
| CN110909784B (en) | Training method and device of image recognition model and electronic equipment | |
| CN113033587B (en) | Image recognition result evaluation method and device, electronic equipment and storage medium | |
| CN113628059A (en) | Associated user identification method and device based on multilayer graph attention network | |
| US20230138302A1 (en) | Multiple scenario-oriented item retrieval method and system | |
| CN111814821A (en) | Deep learning model establishing method, sample processing method and device | |
| CN110674680A (en) | Living body identification method, living body identification device and storage medium | |
| CN111179270A (en) | Image co-segmentation method and device based on attention mechanism | |
| CN111915595A (en) | Image quality evaluation method, and training method and device of image quality evaluation model | |
| CN109886206B (en) | Three-dimensional object identification method and equipment | |
| CN107944363A (en) | Face image processing process, system and server | |
| CN114329004A (en) | Digital fingerprint generation method, digital fingerprint generation device, data push method, data push device and storage medium | |
| CN111860368A (en) | Pedestrian re-identification method, device, equipment and storage medium | |
| CN111814653A (en) | Method, device, equipment and storage medium for detecting abnormal behaviors in video | |
| US20230104345A1 (en) | Image attack detection method and apparatus, and image attack detection model training method and apparatus | |
| CN116630749A (en) | Industrial equipment fault detection method, device, equipment and storage medium | |
| CN116935057A (en) | Target evaluation method, electronic device, and computer-readable storage medium | |
| CN116389711A (en) | Video quality detection method, device, equipment and storage medium | |
| CN113177603B (en) | Training method of classification model, video classification method and related equipment | |
| CN111160219B (en) | Object integrity evaluation method and device, electronic equipment and storage medium | |
| CN114445691A (en) | Model training method and device, electronic equipment and storage medium | |
| CN114120050A (en) | Method, device and equipment for extracting surface ecological data and storage medium | |
| CN114022906A (en) | Pedestrian re-identification method based on multi-level features and attention mechanism |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |