CN110705634A - Heel model identification method and device and storage medium - Google Patents
Heel model identification method and device and storage medium Download PDFInfo
- Publication number
- CN110705634A CN110705634A CN201910930378.4A CN201910930378A CN110705634A CN 110705634 A CN110705634 A CN 110705634A CN 201910930378 A CN201910930378 A CN 201910930378A CN 110705634 A CN110705634 A CN 110705634A
- Authority
- CN
- China
- Prior art keywords
- heel
- image
- candidate
- network
- identifying
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for identifying a heel model, which comprises the following steps: collecting a heel image, and preprocessing the heel image to obtain a heel chromaticity diagram; extracting the features of the heel chromaticity diagram by using a feature extraction network to obtain a feature output diagram; processing the feature output image by using a regional candidate network to obtain a candidate image of the heel; performing pixel-level identification on the candidate image through an output network to obtain the heel height and the heel shape of the candidate image; and identifying the height and the shape of the heel through a heel database to obtain the model of the heel. The identification method, the identification device and the storage medium can improve the speed of identifying the heel, improve the accuracy of identifying the type of the heel and greatly reduce the workload of merchants.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a method and a device for identifying a heel model and a storage medium.
Background
The traditional heel model identification means that a merchant judges the type and model of the heel provided by a customer in a fuzzy way by judging with naked eyes and searching in memory through the heel image or object provided by the customer. Along with the vigorous development of economy, the living standard of people is obviously improved, and the types of shoes are greatly enriched; the market of the corresponding heels also changes greatly, the monthly updated magnitude of the heels of the whole country reaches ten thousand levels according to incomplete statistics, and the market demand cannot be solved through visual identification and memory search. Secondly, the heel image that the customer provided is not standardized through unified collection system and collection standard, in traditional heel discernment, has increased the degree of difficulty of trade company more, has both expended manpower and time to can not guarantee the accuracy of heel model discernment.
Disclosure of Invention
In order to solve the above problems, an object of the present invention is to provide a method, an apparatus, and a storage medium for identifying a heel model, which can improve the speed of identifying the heel, improve the accuracy of identifying the heel model, and greatly reduce the workload of a dealer.
The technical scheme adopted by the invention for solving the problems is as follows: in a first aspect, an embodiment of the present invention provides a method for identifying a heel model, including the following steps:
collecting a heel image, and preprocessing the heel image to obtain a heel chromaticity diagram;
extracting the features of the heel chromaticity diagram by using a feature extraction network to obtain a feature output diagram;
processing the feature output image by using a regional candidate network to obtain a candidate image of the heel;
performing pixel-level identification on the candidate image through an output network to obtain the heel height and the heel shape of the candidate image;
and identifying the height and the shape of the heel through a heel database to obtain the model of the heel.
Further, a heel image is collected, and is preprocessed to obtain a heel chromaticity diagram, and the method comprises the following steps:
acquiring a heel image of the side face of the heel at a horizontal angle within a camera distance range and a camera brightness range;
sharpening the heel image by utilizing high-pass filtering to obtain a sharpened image;
and carrying out highlight denoising processing on the sharpened image by using a bilateral filter to obtain a heel chromaticity diagram.
Further, in the range of the camera shooting distance and the range of the camera shooting brightness, a heel image of the side face of the heel on a horizontal angle is obtained, and the method comprises the following steps:
acquiring the camera shooting distance of the heel, and if the camera shooting distance is not within the range of the camera shooting distance, returning the error information of the camera shooting distance, wherein the range of the camera shooting distance is 10 cm-30 cm;
acquiring the shooting brightness of the heel, and if the shooting brightness is not in the shooting brightness range, returning the shooting brightness error information, wherein the shooting brightness range is that the brightness superposition value of the red, green and blue color channels is not less than 0.4;
and acquiring the camera focal length of the heel and the heel image of the side face of the heel at a horizontal angle.
Further, the feature extraction network includes: a residual network and a feature pyramid network; the residual error network comprises a plurality of residual error blocks, the characteristic pyramid network comprises a plurality of characteristic pyramid network layers, and the rear part of each residual error block is connected with the corresponding characteristic pyramid network layer.
Further, processing the feature output image by using a regional candidate network to obtain a candidate image of the heel, comprising the following steps:
utilizing a regional candidate network to identify the heels of the feature output images to obtain a plurality of candidate regions;
obtaining the confidence coefficient of the candidate region by using a classifier, and screening the candidate region according to the confidence coefficient to obtain a confidence coefficient candidate region;
acquiring the region overlapping degree between the confidence coefficient candidate regions to obtain a region overlapping degree data set of the confidence coefficient candidate regions;
processing the region overlapping degree data set by utilizing non-maximum value inhibition to obtain an optimal candidate region;
and aligning the optimal candidate region by using a bilinear interpolation method to obtain a candidate image.
Further, pixel-level identification is carried out on the candidate image through an output network to obtain the heel height and the heel shape of the candidate image, and the method comprises the following steps:
carrying out pixel-level identification on the candidate image by utilizing a segmentation network to obtain heel pixel points in the candidate image;
obtaining the actual heel height according to the heel pixel points and the camera focal distance;
and classifying the heel pixel points by utilizing a classification network to obtain the heel shape of the candidate image.
Further, the method for identifying the height and the shape of the heel through the heel database to obtain the model of the heel comprises the following steps:
inputting the heel height and the heel shape into a heel database;
and acquiring the heel overlapping degree between the heel height and the heel shape and the data in the heel database, and screening and arranging the heel overlapping degree to obtain a plurality of heel models arranged according to the value of the heel overlapping degree.
In a second aspect, an embodiment of the present invention further provides a heel model identification apparatus, including at least one control processor and a memory, which is in communication connection with the at least one control processor; the memory stores instructions executable by the at least one control processor to enable the at least one control processor to perform a method of heel model identification as defined in any one of the preceding claims.
In a third aspect, the present invention also provides a computer-readable storage medium storing computer-executable instructions for causing a computer to execute a method of identifying a heel model according to any one of the above methods.
The technical scheme provided by the embodiment of the invention at least has the following beneficial effects: collecting a heel image, preprocessing the heel image, and enhancing the definition and resolution of the heel image; the characteristic extraction network is used for extracting the characteristics of the heel chromaticity diagram, so that the speed of extracting the characteristics is improved, and the resolution of the characteristics is enhanced; the feature output image is identified, screened and classified by using the regional candidate network to obtain a normalized candidate image, so that the accuracy of candidate region selection is improved, and the overlapping property between candidate regions is reduced; the candidate images are subjected to pixel level identification through an output network, so that the accuracy of acquiring the height and the shape of the heel is improved; the storage of the heel information through the heel database identifies the height and the shape of the heel of the candidate image, and improves the speed and the efficiency of heel identification.
Drawings
The invention is further illustrated with reference to the following figures and examples.
Fig. 1 is an overall flowchart of one embodiment of a heel model identification method of the present invention.
Detailed Description
Along with the vigorous development of economy, the living standard of people is obviously improved, and the types of shoes are greatly enriched; the corresponding heel market also changes greatly, the monthly updated magnitude of the national heels reaches ten thousand levels, and the market demand cannot be solved through naked eye identification and memory search; and the heel image that the customer provided does not pass through unified collection system and collection standard and standardize, has increaseed the degree of difficulty that the trade company discerned, both consumed manpower and time to can not guarantee the accuracy of heel model discernment.
Based on the method, the device and the storage medium for identifying the heel model, the speed of identifying the heel can be increased, the accuracy rate of identifying the heel model is increased, and the workload of a merchant is greatly reduced.
The embodiments of the present invention will be further explained with reference to the drawings.
Referring to fig. 1, an embodiment of the present invention provides a method for identifying a heel model, including the following steps:
step S100: collecting a heel image, and preprocessing the heel image to obtain a heel chromaticity diagram;
step S200: extracting the features of the heel chromaticity diagram by using a feature extraction network to obtain a feature output diagram;
step S300: processing the feature output image by using a regional candidate network to obtain a candidate image of the heel;
step S400: performing pixel-level identification on the candidate image through an output network to obtain the heel height and the heel shape of the candidate image;
step S500: and identifying the height and the shape of the heel through a heel database to obtain the model of the heel.
In this embodiment, step S100 acquires a heel image, and pre-processes the heel image, so as to enhance the definition and resolution of the heel image; the preprocessing can be sharpening, denoising, brightness adjustment, saturation adjustment and the like. Step S200, feature extraction is carried out on the heel chromaticity diagram by using a feature extraction network, so that the speed of feature extraction is increased, and the resolution of features is enhanced; step S300, the feature output image is identified, screened and classified by using the regional candidate network to obtain a normalized candidate image, so that the accuracy of candidate region selection is improved, and the repeatability among candidate regions is reduced; step S400, the candidate image is subjected to pixel level identification through an output network, so that the heel height and the accuracy of acquiring the heel shape of the candidate image are improved; step S500, the heel information is stored through the heel database, the heel height and the heel shape of the candidate image are identified and compared, and the speed and the efficiency of identifying the heel model are improved.
Further, another embodiment of the present invention further provides a method for identifying a heel model, wherein the method comprises the following steps of collecting a heel image, preprocessing the heel image, and obtaining a heel chromaticity diagram:
step S110: acquiring a heel image of the side face of the heel at a horizontal angle within a camera distance range and a camera brightness range;
step S120: sharpening the heel image by utilizing high-pass filtering to obtain a sharpened image;
step S130: and carrying out highlight denoising processing on the sharpened image by using a bilateral filter to obtain a heel chromaticity diagram.
In this embodiment, step S110 collects images of the heel in the range of the camera distance and the range of the camera brightness, so as to ensure the sharpness of the heel image; the image acquisition is carried out on the horizontal angle, so that the shape difference of the heel is avoided, and the accuracy of the shape of the heel in the heel image is ensured; the image acquisition is carried out on the side surface of the heel, so that the image can acquire more characteristic points of the heel, for example, the image acquisition is carried out on the side surface of the heel of a high-heeled shoe, the gradient of the heel, the thickness of the heel and the like can be obtained from the image of the side surface of the heel of the high-heeled shoe, and the data of the heel cannot be acquired from the front surface, the back surface and the bottom surface of the heel of the high-heeled shoe, so that the image acquisition is carried out on the side surface of the heel, and the accuracy of heel identification can be.
Step S120, the high-pass filtering is a filtering method, the rule is that high-frequency signals can normally pass through, and low-frequency signals lower than a set critical value are blocked and weakened, namely, the high-pass filtering is a filtering process which only has an attenuation effect on frequency components lower than a certain given frequency, allows frequency components above the cut frequency to pass through and has no phase shift; mainly used to eliminate low frequency noise, also called low cut-off filter.
Inputting the heel image into the following formula to carry out high-pass filtering processing:
y(n,m)＝x(n,m)+λz(n,m)；
where x (n, m) is the heel image, y (n, m) is the sharpened image after the high-pass filtering process, and z (n, m) is the correction signal, typically obtained by high-pass filtering x, and λ is a scaling factor for controlling the enhancement effect. The high-frequency part of the heel image is extracted through high-pass filtering, and the high-frequency part is overlapped with the heel image, so that the edge information of the heel image is strengthened, the effect of sharpening the heel image is achieved, and the definition of the heel image is improved.
Step S130, the bilateral filtering is a nonlinear filtering method, is a compromise treatment combining the spatial proximity and the value domain similarity of the image, simultaneously considers the spatial domain information and the maximum diffuse reflection chromaticity similarity, achieves the purpose of edge-preserving and denoising, and has the advantages of simplicity, non-iteration and output weighted combination depending on the neighborhood pixel value. The estimated maximum diffuse reflection chromatic value of the pixel point is used as a weighted combination guide of a value range and a spatial domain to be smooth, denoising and edge-preserving are carried out on the sharpened image, and then the maximum chromatic value of each pixel point is obtained again, namely, the sharpened image is input into the following formula to carry out bilateral filtering treatment:
wherein D is a space domain weight function, R is an estimated maximum diffuse reflection chrominance similarity weight function, p is a pixel point after bilateral filtering processing, q is a pixel point of a sharpened image, and sigma is a color spacemaxIs the maximum diffuse reflectance chromaticity when there is high light,
is the maximum diffuse reflectance chromaticity in the no-highlights state, and Λ max (x) is the estimated maximum diffuse reflectance chromaticity.
After bilateral filtering is carried out on the maximum chromatic value containing the specular reflection pixel points, the chromatic value is reduced, and the maximum chromatic value of the pixel points after filtering is closer to the real maximum diffuse reflection chromatic value. Simultaneously, only the chroma estimation of the pixel points containing diffuse reflectionThe evaluation value will also be reduced by the influence of the pixels containing the specular reflection. Therefore, in order to reduce the influence of the pixel containing the specular component on the chromaticity of the pixel only containing diffuse reflection, the maximum diffuse reflection chromaticity sigma when the pixel contains high light can be comparedmaxAnd estimated maximum diffuse reflectance chromaticity in a no-highlight state
And taking the maximum value as the maximum chromatic value of each pixel:
for sigma in the above formulamaxAnd (4) performing iteration by using bilateral filtering to ensure that the maximum diffuse reflection chromaticity diagram of the same color is smooth. By comparing filtered values after each iteration
And σmaxWhen their difference is less than the threshold at each pixel, the filtered values are considered to converge and the iteration is complete; the threshold value at the pixel can be set according to the actual situation, for example, set to 0.02.
Highlight denoising processing is carried out on the sharpened image through a bilateral filter, and a filtering value is subjected toAnd σmaxIterative comparison processing is carried out to obtain a heel chromaticity diagram containing the maximum chromaticity value of each pixel point of the heel, so that the identification effect of the heel pixel points is improved, and the identification effect of the heel image is improved.
Further, another embodiment of the present invention provides a method for identifying a heel model, wherein the method for obtaining a heel image of a heel side at a horizontal angle in a camera distance range and a camera brightness range includes the following steps:
step S111: acquiring the camera shooting distance of the heel, and if the camera shooting distance is not within the range of the camera shooting distance, returning the error information of the camera shooting distance, wherein the range of the camera shooting distance is 10 cm-30 cm;
step S112: acquiring the shooting brightness of the heel, and if the shooting brightness is not in the shooting brightness range, returning the shooting brightness error information, wherein the shooting brightness range is that the brightness superposition value of the red, green and blue color channels is not less than 0.4;
step S113: and acquiring the camera focal length of the heel and the heel image of the side face of the heel at a horizontal angle.
In this embodiment, the imaging distance range of step S111 is set to 10 cm to 30 cm, so that the size of the heel image acquired by the imaging device is within a certain range, and thus the imaging device can completely acquire the heel image, the specification of the heel image is prevented from being too small or too large, and the sharpness of the heel image acquisition by the imaging device is ensured; when a user collects the heel image, the camera shooting distance is obtained, and when the camera shooting distance is not within the camera shooting distance range, camera shooting distance error information is returned, so that the user can adjust the camera shooting distance through the distance error information, and the accuracy of heel image acquisition is ensured.
In step S112, the photographing brightness of the heel can be represented according to the brightness superposition value of the heel image in the red, green and blue color channels, that is, the photographing brightness calculation formula is:
luminance (RGB) ═ 0.26 red (R) +0.67 green (G) +0.07 blue (B);
wherein, the brightness (RGB) is a brightness superposition value of three color channels of red, green and blue, red (R) is a brightness value of the heel image in the red channel, green (G) is a brightness value of the heel image in the green channel, and blue (B) is a brightness value of the heel image in the blue channel. When the superposition value of the brightness is less than 0.4, the brightness of the heel image is displayed darkly, which is not beneficial to the image processing. Therefore, when the user collects the heel image, the camera brightness is obtained, and when the camera brightness is smaller than 0.4, camera brightness error information is returned, so that the user can adjust the camera brightness through the brightness error information, and the accuracy of obtaining the heel image is ensured.
Step S113, the camera focal length of the heel is automatically adjusted according to the actual shooting environment by the camera equipment, and the calculation of the actual height of the heel is facilitated by acquiring the camera focal length of the heel; acquiring an image of the heel at a horizontal angle, and ensuring the accuracy of the shape of the heel in the heel image; the side face of the heel is subjected to image acquisition, so that the images can acquire more characteristic points of the heel, and the accuracy of heel identification is improved.
Further, another embodiment of the present invention further provides a method for identifying a heel model, wherein the feature extraction network includes: a residual network and a feature pyramid network; the residual error network comprises a plurality of residual error blocks, the characteristic pyramid network comprises a plurality of characteristic pyramid network layers, and the rear part of each residual error block is connected with the corresponding characteristic pyramid network layer.
In the embodiment, the residual error network is characterized by being easy to optimize, and the accuracy can be improved by increasing the equivalent depth, and the gradient disappearance problem caused by increasing the depth in the deep neural network is relieved because the internal residual error block uses jump connection; the characteristic pyramid is used for detecting and identifying objects with different scales, and the characteristic pyramid with marginal extra loss is constructed by utilizing the inherent multi-scale pyramid hierarchical structure of the deep convolutional network, so that the network has a transversely-connected top-down architecture, and a high-level semantic characteristic diagram can be constructed on all scales. The residual error network comprises a plurality of residual error blocks, and a characteristic pyramid network layer can be connected behind any residual error block, namely, in the plurality of residual error blocks, at most, a characteristic pyramid network layer can be connected behind each residual error block. The heel chromaticity diagram is input into the feature extraction network for feature extraction, so that the speed of heel feature extraction is improved, and the resolution of heel features is enhanced. The number of layers of the feature pyramid network is not limited, and the feature pyramid network is set according to the number of actual residual blocks.
Further, another embodiment of the present invention further provides a method for identifying a heel model, wherein the method for processing the feature output map by using a regional candidate network to obtain a candidate image of a heel includes the following steps:
step S310: utilizing a regional candidate network to identify the heels of the feature output images to obtain a plurality of candidate regions;
step S320: obtaining the confidence coefficient of the candidate region by using a classifier, and screening the candidate region according to the confidence coefficient to obtain a confidence coefficient candidate region;
step S330: acquiring the region overlapping degree between the confidence coefficient candidate regions to obtain a region overlapping degree data set of the confidence coefficient candidate regions;
step S340: processing the region overlapping degree data set by utilizing non-maximum value inhibition to obtain an optimal candidate region;
step S350: and aligning the optimal candidate region by using a bilinear interpolation method to obtain a candidate image.
In this embodiment, in step S310, the area candidate network traverses all points on the feature output map by using a sliding window, and determines all interested areas on the feature output map to obtain a plurality of candidate areas. Step S320, calculating the confidence coefficient of the candidate regions by using a classifier, screening out a plurality of candidate regions with the highest confidence coefficient according to the confidence coefficient, and marking as confidence coefficient candidate regions; step S330 obtains the region overlapping degree between the confidence candidate regions to obtain a region overlapping degree data set of the confidence candidate regions, that is, the region overlapping degree data set of each confidence candidate region includes data of the region overlapping degree between the candidate region and other candidate regions.
Step S340 is to suppress the non-maximum element, i.e. to select local maximum search, where the local representation is a neighborhood, and the neighborhood has two variable parameters, namely, the dimension of the neighborhood and the size of the neighborhood. The specific searching step for the data group of the region overlapping degree comprises the following steps: marking the confidence coefficient candidate region as a first confidence coefficient candidate region from the confidence coefficient candidate region with the maximum confidence coefficient, and screening out a first class of confidence coefficient candidate regions with the region overlapping degree value not greater than a threshold value from a region overlapping degree data set of the first confidence coefficient candidate region; selecting a second confidence coefficient candidate region with the maximum region confidence coefficient from the first confidence coefficient candidate regions, and screening a third confidence coefficient candidate region with a region overlap value not greater than a threshold value in a region overlap degree data set of the second confidence coefficient candidate region; and continuing the screening until the maximum confidence coefficient candidate region in all the Nth confidence coefficient candidate regions is selected and merged to obtain the optimal candidate region of the heel, so that the accuracy of selecting the candidate regions is improved, and the overlapping property between the candidate regions is reduced. Wherein the range of the N-1 type confidence degree candidate region comprises the N type confidence degree candidate region.
For example, the confidence candidate regions include A, B, C, D, E and F, and the region confidence of the confidence candidate region is a < B < C < D < E < F, first extracting F with the highest region confidence, and marking F as the first confidence candidate region; in the region overlapping degree data set of F, screening out confidence coefficient candidate regions with the region overlapping degree not larger than a threshold, and if the region overlapping degree values of A, F, B, F, C and F are not larger than the threshold, screening out A, B and C, and recording first class confidence coefficient candidate regions; since the region confidence coefficient of C is greater than the region confidence coefficients of A and B, marking C as a second confidence coefficient candidate region, and screening the region overlapping degrees of C and A, C only according to the region overlapping degree of C in the region overlapping degree data set of C to obtain A with the region overlapping degree value not greater than a threshold value; then the optimal candidate region is a merge region of F, C, A.
The bilinear interpolation in step S350 is a linear interpolation extension of an interpolation function of two variables, and the core idea is to perform linear interpolation in two directions respectively. Selecting four pixel points with fixed positions in the optimal candidate region, and performing bilinear interpolation on the four pixel points with fixed positions, wherein the bilinear interpolation process comprises the following steps: for each pixel point at a fixed position, four heel pixel points adjacent to the pixel point are selected in the optimal candidate area, linear interpolation in the horizontal direction and the vertical direction is carried out on the four heel pixel points, namely, the corresponding weight is determined according to the distance between the pixel point at the fixed position and the four heel pixel points, and therefore the interpolation position of the pixel point at the fixed position is calculated. The principle of bilinear interpolation is: taking the distance from the pixel point at the fixed position to the adjacent nearest four heel pixel points as a reference weight, and obtaining the interpolation position of the pixel point at the fixed position through twice linear interpolation; and aligning the optimal candidate area according to the interpolation positions of the pixel points at the four fixed positions to obtain a normalized candidate image, so that the accuracy of heel image identification is improved.
Further, another embodiment of the present invention further provides a method for identifying a heel model, wherein the method for identifying a heel model of a candidate image by using an output network performs pixel-level identification on the candidate image to obtain a heel height and a heel shape of the candidate image, and comprises the following steps:
step S410: carrying out pixel-level identification on the candidate image by utilizing a segmentation network to obtain heel pixel points in the candidate image;
step S420: obtaining the actual heel height according to the heel pixel points and the camera focal distance;
step S430: and classifying the heel pixel points by utilizing a classification network to obtain the heel shape of the candidate image.
In this embodiment, there are various types of the split networks in step S410, and the common split networks include: FCN, UNet, SegNet, deep Lab, etc., the segmentation network is to carry out pixel level identification and classification to the candidate image, and the types of the pixel points of the obtained candidate image are as follows: and simultaneously, screening pixel points of the candidate image to obtain heel pixel points.
Step S420 may calculate the height of the heel in the candidate image according to the position of the pixel point of the heel, and may calculate the actual height of the heel by the following formula:
wherein f is the camera focal length, H is the heel height of the candidate image, D is the camera distance, and H is the actual heel height.
The classification network of step S430 has various types, and the common classification networks include: LeNet-5, AlexNet, ZFNET, VGGNet, GoogLeNet, ResNet, etc.; the classification network mainly utilizes operations such as convolution, parameter sharing, pooling and the like to extract features, and uses the fully-connected neural network to classify and identify the features, so that a large amount of calculation among data is reduced. And classifying the heel pixel points by using a classification network to obtain the shape formed by all the pixel points of the same heel, namely obtaining the heel shape of the candidate image.
Further, another embodiment of the present invention further provides a method for identifying a heel model, wherein identifying the height of the heel and the shape of the heel through a heel database to obtain the model of the heel comprises the following steps:
step S510: inputting the heel height and the heel shape into a heel database;
step S520: and acquiring the heel overlapping degree between the heel height and the heel shape and the data in the heel database, and screening and arranging the heel overlapping degree to obtain a plurality of heel models arranged according to the value of the heel overlapping degree.
In this embodiment, in step S520, a large amount of heel information is stored in the heel database, and for the same heel, the heel information includes a real heel height and a real heel shape; through calculating the height and the shape of the heel, the heel overlapping degree between the real heel height and the real heel shape in the heel database is sorted according to the size of the heel overlapping degree, a plurality of heel models with larger heel overlapping degree are screened out, and the plurality of heel models are output according to the size of the overlapping degree value, so that a user can obtain a plurality of heel models with the highest similarity in a heel image, the accuracy of heel model identification is improved, and the workload of a merchant is greatly reduced. Wherein, a plurality of heel models with larger heel overlapping degree can be set as ten heel models with the largest heel overlapping degree.
In addition, referring to fig. 1, another embodiment of the present invention further provides a method for identifying a heel model, including the following steps:
step S111: acquiring the camera shooting distance of the heel, and if the camera shooting distance is not within the range of the camera shooting distance, returning the error information of the camera shooting distance, wherein the range of the camera shooting distance is 10 cm-30 cm;
step S112: acquiring the shooting brightness of the heel, and if the shooting brightness is not in the shooting brightness range, returning the shooting brightness error information, wherein the shooting brightness range is that the brightness superposition value of the red, green and blue color channels is not less than 0.4;
step S113: acquiring a camera focal length of a heel and a heel image of the side face of the heel at a horizontal angle;
step S120: sharpening the heel image by utilizing high-pass filtering to obtain a sharpened image;
step S130: highlight denoising processing is carried out on the sharpened image by using a bilateral filter to obtain a heel chromaticity diagram;
step S200: extracting the features of the heel chromaticity diagram by using a feature extraction network to obtain a feature output diagram;
step S310: utilizing a regional candidate network to identify the heels of the feature output images to obtain a plurality of candidate regions;
step S320: obtaining the confidence coefficient of the candidate region by using a classifier, and screening the candidate region according to the confidence coefficient to obtain a confidence coefficient candidate region;
step S330: acquiring the region overlapping degree between the confidence coefficient candidate regions to obtain a region overlapping degree data set of the confidence coefficient candidate regions;
step S340: processing the region overlapping degree data set by utilizing non-maximum value inhibition to obtain an optimal candidate region;
step S350: aligning the optimal candidate area by using a bilinear interpolation method to obtain a candidate image;
step S410: carrying out pixel-level identification on the candidate image by utilizing a segmentation network to obtain heel pixel points in the candidate image;
step S420: obtaining the actual heel height according to the heel pixel points and the camera focal distance;
step S430: classifying the heel pixel points by utilizing a classification network to obtain the heel shape of the candidate image;
step S510: inputting the heel height and the heel shape into a heel database;
step S520: and acquiring the heel overlapping degree between the heel height and the heel shape and the data in the heel database, and screening and arranging the heel overlapping degree to obtain a plurality of heel models arranged according to the value of the heel overlapping degree.
In the embodiment, a heel image is acquired, and is preprocessed, so that the definition and the resolution of the heel image are enhanced; the characteristic extraction network is used for extracting the characteristics of the heel chromaticity diagram, so that the speed of extracting the characteristics is improved, and the resolution of the characteristics is enhanced; the feature output image is identified, screened and classified by using the regional candidate network to obtain a normalized candidate image, so that the accuracy of candidate region selection is improved, and the overlapping property between candidate regions is reduced; the candidate images are subjected to pixel level identification through an output network, so that the accuracy of acquiring the height and the shape of the heel is improved; the storage of the heel information through the heel database identifies the height and the shape of the heel of the candidate image, and improves the speed and the efficiency of heel identification.
In addition, another embodiment of the invention also provides a heel model identification device, which comprises at least one control processor and a memory used for being connected with the at least one control processor in a communication way; the memory stores instructions executable by the at least one control processor to enable the at least one control processor to perform a method of heel model identification as defined in any one of the preceding claims.
In this embodiment, the identification means includes: one or more control processors and memory, which may be connected by a bus or otherwise.
The memory, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs, non-transitory computer executable programs, and modules, such as program instructions/modules corresponding to the identification methods in the embodiments of the present invention. The control processor executes various functional applications and data processing of the recognition device by running non-transitory software programs, instructions and modules stored in the memory, i.e. implementing the recognition method of the above-described method embodiments.
The memory may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to use of the identification device, and the like. Further, the memory may include high speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory optionally includes memory located remotely from the control processor, and these remote memories may be connected to the identification appliance via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The one or more modules are stored in the memory and, when executed by the one or more control processors, perform the identification method of the above-described method embodiments, e.g., perform the functions of the above-described identification method steps S100 to S500, S110 to S130, S111 to S113, S310 to S350, S410 to S430, and S510 to S520.
Embodiments of the present invention also provide a computer-readable storage medium storing computer-executable instructions, which are executed by one or more control processors, for example, a control processor, and can enable the one or more control processors to execute the identification method in the above method embodiments, for example, execute the functions of the above-described method steps S100 to S500, S110 to S130, S111 to S113, S310 to S350, S410 to S430, and S510 to S520.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, may be located in one place, or may be distributed over a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment.
Through the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by software plus a general hardware platform. Those skilled in the art will appreciate that all or part of the processes of the methods of the above embodiments may be implemented by hardware related to instructions of a computer program, which may be stored in a computer readable storage medium, and when executed, may include the processes 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 (RAM), or the like.
While the preferred embodiments of the present invention have been described in detail, it will be understood by those skilled in the art that the foregoing and various other changes, omissions and deviations in the form and detail thereof may be made without departing from the scope of this invention.
Claims (9)
1. A method for identifying a heel model is characterized by comprising the following steps:
collecting a heel image, and preprocessing the heel image to obtain a heel chromaticity diagram;
extracting the features of the heel chromaticity diagram by using a feature extraction network to obtain a feature output diagram;
processing the feature output image by using a regional candidate network to obtain a candidate image of the heel;
performing pixel-level identification on the candidate image through an output network to obtain the heel height and the heel shape of the candidate image;
and identifying the height and the shape of the heel through a heel database to obtain the model of the heel.
2. The method for identifying the heel model according to claim 1, wherein: the method comprises the following steps of collecting a heel image, preprocessing the heel image to obtain a heel chromaticity diagram, and comprises the following steps:
acquiring a heel image of the side face of the heel at a horizontal angle within a camera distance range and a camera brightness range;
sharpening the heel image by utilizing high-pass filtering to obtain a sharpened image;
and carrying out highlight denoising processing on the sharpened image by using a bilateral filter to obtain a heel chromaticity diagram.
3. The method for identifying the heel model according to claim 2, wherein: in the range of the camera shooting distance and the range of the camera shooting brightness, a heel image of the side face of the heel on a horizontal angle is acquired, and the method comprises the following steps:
acquiring the camera shooting distance of the heel, and if the camera shooting distance is not within the range of the camera shooting distance, returning the error information of the camera shooting distance, wherein the range of the camera shooting distance is 10 cm-30 cm;
acquiring the shooting brightness of the heel, and if the shooting brightness is not in the shooting brightness range, returning the shooting brightness error information, wherein the shooting brightness range is that the brightness superposition value of the red, green and blue color channels is not less than 0.4;
and acquiring the camera focal length of the heel and the heel image of the side face of the heel at a horizontal angle.
4. The method for identifying the heel model according to claim 1, wherein: the feature extraction network includes: a residual network and a feature pyramid network; the residual error network comprises a plurality of residual error blocks, the characteristic pyramid network comprises a plurality of characteristic pyramid network layers, and the rear part of each residual error block is connected with the corresponding characteristic pyramid network layer.
5. The method for identifying the heel model according to claim 1, wherein: processing the feature output image by using a regional candidate network to obtain a candidate image of the heel, comprising the following steps of:
utilizing a regional candidate network to identify the heels of the feature output images to obtain a plurality of candidate regions;
obtaining the confidence coefficient of the candidate region by using a classifier, and screening the candidate region according to the confidence coefficient to obtain a confidence coefficient candidate region;
acquiring the region overlapping degree between the confidence coefficient candidate regions to obtain a region overlapping degree data set of the confidence coefficient candidate regions;
processing the region overlapping degree data set by utilizing non-maximum value inhibition to obtain an optimal candidate region;
and aligning the optimal candidate region by using a bilinear interpolation method to obtain a candidate image.
6. A method of identifying a heel type according to claim 3, wherein: the pixel-level identification is carried out on the candidate image through an output network to obtain the heel height and the heel shape of the candidate image, and the method comprises the following steps:
carrying out pixel-level identification on the candidate image by utilizing a segmentation network to obtain heel pixel points in the candidate image;
obtaining the actual heel height according to the heel pixel points and the camera focal distance;
and classifying the heel pixel points by utilizing a classification network to obtain the heel shape of the candidate image.
7. The method for identifying the heel model according to claim 1, wherein: identifying the height and the shape of the heel through a heel database to obtain the model of the heel, comprising the following steps:
inputting the heel height and the heel shape into a heel database;
and acquiring the heel overlapping degree between the heel height and the heel shape and the data in the heel database, and screening and arranging the heel overlapping degree to obtain a plurality of heel models arranged according to the value of the heel overlapping degree.
8. A heel model identification device comprising at least one control processor and a memory for communicative connection with said at least one control processor; the memory stores instructions executable by the at least one control processor to enable the at least one control processor to perform the heel model identification method of any one of claims 1-7.
9. A computer-readable storage medium storing computer-executable instructions for causing a computer to perform the heel model identification method according to any one of claims 1 to 7.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910930378.4A CN110705634B (en) | 2019-09-29 | 2019-09-29 | Heel model identification method and device and storage medium |
| PCT/CN2020/112536 WO2021057395A1 (en) | 2019-09-29 | 2020-08-31 | Heel type identification method, device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910930378.4A CN110705634B (en) | 2019-09-29 | 2019-09-29 | Heel model identification method and device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110705634A true CN110705634A (en) | 2020-01-17 |
| CN110705634B CN110705634B (en) | 2023-02-28 |
Family
ID=69197862
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910930378.4A Active CN110705634B (en) | 2019-09-29 | 2019-09-29 | Heel model identification method and device and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110705634B (en) |
| WO (1) | WO2021057395A1 (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021057395A1 (en) * | 2019-09-29 | 2021-04-01 | 五邑大学 | Heel type identification method, device, and storage medium |
| CN113496221A (en) * | 2021-09-08 | 2021-10-12 | 湖南大学 | Point supervision remote sensing image semantic segmentation method and system based on depth bilateral filtering |
| CN114619663A (en) * | 2022-03-22 | 2022-06-14 | 芜湖风雪橡胶有限公司 | Rubber shoe sole antiskid line pressing forming equipment and pressing method |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113221795B (en) * | 2021-05-24 | 2024-05-14 | 大连恒锐科技股份有限公司 | Method and device for extracting, fusing and comparing shoe pattern features in video |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102324020A (en) * | 2011-09-02 | 2012-01-18 | 北京新媒传信科技有限公司 | The recognition methods of area of skin color of human body and device |
| CN107346413A (en) * | 2017-05-16 | 2017-11-14 | 北京建筑大学 | Traffic sign recognition method and system in a kind of streetscape image |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104680163A (en) * | 2015-02-10 | 2015-06-03 | 柳州市金旭节能科技有限公司 | Licence plate recognition system |
| WO2017120794A1 (en) * | 2016-01-13 | 2017-07-20 | 北京大学深圳研究生院 | Image matching method and apparatus |
| CN106875381B (en) * | 2017-01-17 | 2020-04-28 | 同济大学 | Mobile phone shell defect detection method based on deep learning |
| CN109916308B (en) * | 2019-01-14 | 2021-08-24 | 佛山市南海区广工大数控装备协同创新研究院 | Information acquisition method and system for shoe sole |
| CN109902749A (en) * | 2019-03-04 | 2019-06-18 | 沈阳建筑大学 | A kind of print recognition methods of shoes and system |
| CN110705634B (en) * | 2019-09-29 | 2023-02-28 | 五邑大学 | Heel model identification method and device and storage medium |
-
2019
- 2019-09-29 CN CN201910930378.4A patent/CN110705634B/en active Active
-
2020
- 2020-08-31 WO PCT/CN2020/112536 patent/WO2021057395A1/en active Application Filing
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102324020A (en) * | 2011-09-02 | 2012-01-18 | 北京新媒传信科技有限公司 | The recognition methods of area of skin color of human body and device |
| CN107346413A (en) * | 2017-05-16 | 2017-11-14 | 北京建筑大学 | Traffic sign recognition method and system in a kind of streetscape image |
Non-Patent Citations (1)
| Title |
|---|
| 宋红等: "基于人脸检测与跟踪的智能监控系统", 《北京理工大学学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021057395A1 (en) * | 2019-09-29 | 2021-04-01 | 五邑大学 | Heel type identification method, device, and storage medium |
| CN113496221A (en) * | 2021-09-08 | 2021-10-12 | 湖南大学 | Point supervision remote sensing image semantic segmentation method and system based on depth bilateral filtering |
| CN113496221B (en) * | 2021-09-08 | 2022-02-01 | 湖南大学 | Point supervision remote sensing image semantic segmentation method and system based on depth bilateral filtering |
| CN114619663A (en) * | 2022-03-22 | 2022-06-14 | 芜湖风雪橡胶有限公司 | Rubber shoe sole antiskid line pressing forming equipment and pressing method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110705634B (en) | 2023-02-28 |
| WO2021057395A1 (en) | 2021-04-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110705634B (en) | Heel model identification method and device and storage medium | |
| EP3455782B1 (en) | System and method for detecting plant diseases | |
| Santra et al. | Learning a patch quality comparator for single image dehazing | |
| Park et al. | Single image dehazing with image entropy and information fidelity | |
| CN108389224B (en) | Image processing method and device, electronic equipment and storage medium | |
| CN110309781B (en) | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion | |
| KR20180065889A (en) | Method and apparatus for detecting target | |
| TW200834459A (en) | Video object segmentation method applied for rainy situations | |
| CN114118144A (en) | Anti-interference accurate aerial remote sensing image shadow detection method | |
| US20130342694A1 (en) | Method and system for use of intrinsic images in an automotive driver-vehicle-assistance device | |
| CN108377374A (en) | Method and system for generating depth information related to an image | |
| KR101833943B1 (en) | Method and system for extracting and searching highlight image | |
| CN105701813A (en) | Significance detection method of light field image | |
| Wang et al. | An efficient method for image dehazing | |
| Singh et al. | Visibility enhancement and dehazing: Research contribution challenges and direction | |
| CN116883303A (en) | Infrared and visible light image fusion method based on characteristic difference compensation and fusion | |
| Wang et al. | Haze removal algorithm based on single-images with chromatic properties | |
| Zhang et al. | Deep joint neural model for single image haze removal and color correction | |
| JP5914046B2 (en) | Image processing apparatus and image processing method | |
| JP6334281B2 (en) | Forest phase analysis apparatus, forest phase analysis method and program | |
| US8428352B1 (en) | Post processing for improved generation of intrinsic images | |
| Prasenan et al. | A Study of Underwater Image Pre-processing and Techniques | |
| Nair et al. | Benchmarking single image dehazing methods | |
| JP6218678B2 (en) | Forest phase analysis apparatus, forest phase analysis method and program | |
| WO2013070349A1 (en) | Post processing for improved generation of intrinsic images |
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 |