WO2019024610A1 - Procédé de reconnaissance d'image, appareil, dispositif informatique et support d'informations lisible - Google Patents

Procédé de reconnaissance d'image, appareil, dispositif informatique et support d'informations lisible Download PDF

Info

Publication number
WO2019024610A1
WO2019024610A1 PCT/CN2018/090970 CN2018090970W WO2019024610A1 WO 2019024610 A1 WO2019024610 A1 WO 2019024610A1 CN 2018090970 W CN2018090970 W CN 2018090970W WO 2019024610 A1 WO2019024610 A1 WO 2019024610A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
feature
preset
matching degree
shape
Prior art date
Application number
PCT/CN2018/090970
Other languages
English (en)
Chinese (zh)
Inventor
郭浒生
Original Assignee
合肥美的智能科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 合肥美的智能科技有限公司 filed Critical 合肥美的智能科技有限公司
Publication of WO2019024610A1 publication Critical patent/WO2019024610A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/68—Food, e.g. fruit or vegetables

Definitions

  • the present invention relates to the field of image recognition technology, and in particular to an image recognition method, an image recognition device, a computer device, and a computer readable storage medium.
  • image recognition of food materials is generally implemented based on a neural network model for machine training learning to mark a large number of sample food images, and the following technical defects exist:
  • the present invention aims to solve at least one of the technical problems existing in the prior art or related art.
  • Another object of the present invention is to provide an image recognition apparatus.
  • the technical solution of the first aspect of the present invention provides an image recognition method, including: determining a shape feature of an image to be recognized; extracting a texture feature and a color feature of the image to be recognized; according to the first preset image library The image feature weight table, determining the weight of each texture feature of the image to be identified and the weight of each color feature; the filtering weight is greater than the texture feature and the color feature of the preset weight to form the texture feature set and the color feature set respectively; Determining, according to the shape feature, the texture feature set and the color feature set, a first evolution image corresponding to the image to be identified; comparing a matching degree of the first evolution image with any one of the second preset image library, and determining a matching image of the image to be recognized .
  • the feature of the image to be recognized is extracted, by using the image feature weight table in the first preset image library. Determining the weight of each texture feature of the image to be identified and the weight of each color feature, and then filtering the texture features and color features that are greater than the preset weights to form the texture feature set and the color feature set respectively, further optimizing the waiting for Identifying the features extracted from the image is beneficial to improve the recognition accuracy of the image to be recognized.
  • the first evolution image corresponding to the image to be identified is determined according to the shape feature, the texture feature set and the color feature set, and the image to be recognized is further optimized.
  • the main image feature of the image to be recognized is highlighted in the first evolution image, which is beneficial to reduce the matching difficulty by comparing any image in the first evolution image and the second preset image library.
  • Matching degree determine the matching image of the image to be identified, and realize The recognition of the image to be recognized improves the accuracy of the image recognition, and the number of sample images in the second preset image library only needs to include a small number of sample images containing the main image features to realize the recognition of the image to be recognized, and the sample image is reduced. The number saves hardware resources and reduces the difficulty of machine training.
  • the area suggestion frame is divided by the area suggestion network, and the image in the area suggestion frame is the image to be recognized, and the image to be recognized includes the image of the food material and the background image of the position, and is other
  • the image of the object is occluded, and the method of the present invention can screen the image features of the food material, and then obtain the first evolution image, so that the image of the sample material in the second preset image library can be matched, and then the matching degree is compared. To achieve image recognition of the image to be recognized.
  • the preset weight is 0.4
  • the first preset image library is a user's preference library
  • the image feature weight table in the first preset image library is based on any image feature in the first preset image library in all image features.
  • the weight of the image to be identified is determined by the gray level co-occurrence matrix algorithm, and the color feature of the image to be recognized can be extracted by a histogram algorithm.
  • the size of the image to be identified is M ⁇ N
  • the gray level is L
  • G ⁇ 0, 1, 2, ..., L-1 ⁇
  • f(x, y) is the gray of the pixel at the coordinates (x, y).
  • Degree level a co-occurrence matrix of each image is an L ⁇ L matrix T[t ij ] L ⁇ L
  • the elements in T are the spatial relationship of the gray level of the image, Extracted texture features p ij ,
  • the number of values, nk is the number of pixels in the image with a pixel feature value of k, and N is the total number of image pixels.
  • determining a shape feature of the image to be identified includes: extracting a corner feature of the image to be recognized; calculating a matching degree of the corner feature with any shape in the preset shape library to determine a maximum matching degree. Determining whether the maximum matching degree is greater than the first preset matching degree threshold; if it is determined that the maximum matching degree is greater than the first preset matching degree threshold, determining that the shape corresponding to the maximum matching degree is the shape feature of the image to be identified; if determining the maximum matching degree If it is not greater than the first preset matching degree threshold, the corner feature of the image to be recognized is continuously extracted.
  • the maximum matching degree is determined by extracting the corner feature of the image to be recognized, and calculating the matching degree of the corner feature with any shape in the preset shape library, and then the maximum matching degree is greater than the first preset matching.
  • the threshold is determined, the shape corresponding to the maximum matching degree is determined as the shape feature of the image to be recognized, the shape feature of the image to be recognized is determined, and the accuracy of the shape feature is determined, which is beneficial to further improve the accuracy of image recognition. .
  • the first preset matching degree threshold is 0.5-0.8, and the corner feature extraction of the image to be identified can be implemented by a Harris corner feature extraction algorithm.
  • the corner response value D is greater than the set threshold and is the local maximum in the neighborhood of the point, the point is extracted as a corner feature.
  • determining, according to the shape feature, the texture feature set, and the color feature set, the first evolution image corresponding to the image to be identified including: constructing a shape space according to the shape feature; randomly selecting the texture feature set The color features in the texture feature and the color feature set are respectively set in the shape space to form a second evolution image; determining whether the second evolution image satisfies the first preset condition; and if the second evolution image satisfies the first pre-determination Setting a condition, determining that the second evolution image is the first evolution image; if it is determined that the second evolution image does not satisfy the first preset condition, then re-selecting the texture feature in the texture feature set and the color feature in the color feature set are sequentially And being disposed in the shape space to form a new second evolution image, wherein the first preset condition is that the maximum matching degree between the second evolution image and the image in the second preset image library is not less than a second preset matching degree threshold And/or
  • the shape space is constructed according to the shape feature, and then the texture feature in the texture feature set and the color feature in the color feature set are randomly selected and sequentially set in the shape space to form a second evolution image, and the second evolution image is improved.
  • the diversity of the evolved image is advantageous for improving the efficiency of image recognition, and is also beneficial for reducing the number of sample images, by judging whether the second evolved image satisfies the first preset condition, and when the second evolved image satisfies the first preset condition, Determining that the second evolution image is the first evolution image, and continuing to form a new second evolution image when the second evolution image does not satisfy the first preset condition, further reducing the first evolution image and the second preset image library
  • the image matching difficulty increases the accuracy of the recognition of the image to be recognized.
  • the second preset matching degree threshold is 0.6-0.9, and the first preset number of thresholds is 8000-11000.
  • determining, according to the shape feature, the texture feature set, and the color feature set, the first evolution image corresponding to the image to be identified including: determining the shape feature, the texture feature set, and the color feature set to initialize a group; mutating, intersecting, and selecting operations are sequentially performed on the initialization group to form an evolutionary group; determining whether the evolutionary group satisfies a second predetermined condition; if the evolutionary group is determined to satisfy the second predetermined condition, determining an image corresponding to the evolved group as The first evolutionary image; if it is determined that the evolutionary group does not satisfy the second predetermined condition, the evolutionary group is used as an initialization group, and the mutation, intersection, and selection operations are sequentially performed, wherein the second preset condition is mutation, intersection, and selection operation.
  • the number of times is not less than the second preset number threshold and/or the maximum matching degree of the image corresponding to the evolution group and the image in the preset image library is not less than the third preset matching degree threshold
  • the second preset number threshold is 8000-11000
  • the third preset matching threshold is 0.6-0.9.
  • the mutation, crossover, and selection operations can be implemented by a differential evolution algorithm.
  • the method further includes: adding the matching image to the first A preset image library and update the image feature weight table.
  • the association between the image feature weight table and the user's preference is realized, and the image to be identified is filtered according to the user's preference.
  • the realization of the image features further improves the accuracy of image recognition of the image to be recognized.
  • the technical solution of the second aspect of the present invention provides an image recognition apparatus, including: a determining unit, configured to determine a shape feature of an image to be recognized; an extracting unit, configured to extract a texture feature and a color feature of the image to be identified; The method is further configured to: determine, according to an image feature weight table in the first preset image library, a weight of each texture feature of the image to be identified and a weight of each color feature; and a screening unit, configured to filter a texture whose weight is greater than a preset weight a feature and a color feature to respectively form a texture feature set and a color feature set; the determining unit is further configured to: determine a first evolution image corresponding to the image to be identified according to the shape feature, the texture feature set, and the color feature set; and the comparing unit is configured to: Comparing the matching degree of the first evolved image with any of the second preset image libraries to determine a matching image of the image to be recognized.
  • the feature of the image to be recognized is extracted, by using the image feature weight table in the first preset image library. Determining the weight of each texture feature of the image to be identified and the weight of each color feature, and then filtering the texture features and color features that are greater than the preset weights to form the texture feature set and the color feature set respectively, further optimizing the waiting for Identifying the features extracted from the image is beneficial to improve the recognition accuracy of the image to be recognized.
  • the first evolution image corresponding to the image to be identified is determined according to the shape feature, the texture feature set and the color feature set, and the image to be recognized is further optimized.
  • the main image feature of the image to be recognized is highlighted in the first evolution image, which is beneficial to reduce the matching difficulty by comparing any image in the first evolution image and the second preset image library.
  • Matching degree determine the matching image of the image to be identified, and realize The recognition of the image to be recognized improves the accuracy of the image recognition, and the number of sample images in the second preset image library only needs to include a small number of sample images containing the main image features to realize the recognition of the image to be recognized, and the sample image is reduced. The number saves hardware resources and reduces the difficulty of machine training.
  • the area suggestion frame is divided by the area suggestion network, and the image in the area suggestion frame is the image to be recognized, and the image to be recognized includes the image of the food material and the background image of the position, and is other
  • the image of the object is occluded, and the method of the present invention can screen the image features of the food material, and then obtain the first evolution image, so that the image of the sample material in the second preset image library can be matched, and then the matching degree is compared. To achieve image recognition of the image to be recognized.
  • the preset weight is 0.4
  • the first preset image library is a user's preference library
  • the image feature weight table in the first preset image library is based on any image feature in the first preset image library in all image features.
  • the weight of the image to be identified is determined by the gray level co-occurrence matrix algorithm, and the color feature of the image to be recognized can be extracted by a histogram algorithm.
  • the size of the image to be identified is M ⁇ N
  • the gray level is L
  • G ⁇ 0, 1, 2, ..., L-1 ⁇
  • f(x, y) is the gray of the pixel at the coordinates (x, y).
  • Degree level a co-occurrence matrix of each image is an L ⁇ L matrix T[t ij ] L ⁇ L
  • the elements in T are the spatial relationship of the gray level of the image, Extracted texture features p ij ,
  • the number of values, nk is the number of pixels in the image with a pixel feature value of k, and N is the total number of image pixels.
  • the extracting unit is further configured to: extract a corner feature of the image to be recognized;
  • the image recognition device further includes: a calculating unit, configured to calculate a corner feature and any shape in the preset shape library Matching degree to determine a maximum matching degree;
  • the first determining unit is configured to determine whether the maximum matching degree is greater than the first preset matching degree threshold;
  • the determining unit is further configured to: determine, in the first determining unit, that the maximum matching degree is greater than the first preset
  • the shape corresponding to the maximum matching degree is determined as the shape feature of the image to be identified;
  • the extracting unit is further configured to: when the first determining unit determines that the maximum matching degree is not greater than the first preset matching degree threshold, continue to extract the to-be-identified The corner feature of the image.
  • the maximum matching degree is determined by extracting the corner feature of the image to be recognized, and calculating the matching degree of the corner feature with any shape in the preset shape library, and then the maximum matching degree is greater than the first preset matching.
  • the threshold is determined, the shape corresponding to the maximum matching degree is determined as the shape feature of the image to be recognized, the shape feature of the image to be recognized is determined, and the accuracy of the shape feature is determined, which is beneficial to further improve the accuracy of image recognition. .
  • the first preset matching degree threshold is 0.5-0.8, and the corner feature extraction of the image to be identified can be implemented by a Harris corner feature extraction algorithm.
  • the corner response value D is greater than the set threshold and is the local maximum in the neighborhood of the point, the point is extracted as a corner feature.
  • the method further includes: a building unit configured to construct a shape space according to the shape feature; and a selecting unit configured to randomly select the texture feature in the texture feature set and the color feature in the color feature set
  • the second determining unit is configured to determine whether the second evolved image satisfies the first preset condition, and the determining unit is further configured to determine the second evolution in the second determining unit.
  • the selecting unit is further configured to: re-randomly select the texture feature when the second determining unit determines that the second evolved image does not satisfy the first preset condition
  • the texture features in the collection and the color features in the color feature set are respectively disposed in the shape space to form a new second evolution image, wherein the first preset condition is in the second evolution image and the second preset image library.
  • the maximum matching degree of the image is not less than the second preset matching degree threshold and/or the number of judgments is not less than the first preset number of times threshold.
  • the shape space is constructed according to the shape feature, and then the texture feature in the texture feature set and the color feature in the color feature set are randomly selected and sequentially set in the shape space to form a second evolution image, and the second evolution image is improved.
  • the diversity of the evolved image is advantageous for improving the efficiency of image recognition, and is also beneficial for reducing the number of sample images, by judging whether the second evolved image satisfies the first preset condition, and when the second evolved image satisfies the first preset condition, Determining that the second evolution image is the first evolution image, and continuing to form a new second evolution image when the second evolution image does not satisfy the first preset condition, further reducing the first evolution image and the second preset image library
  • the image matching difficulty increases the accuracy of the recognition of the image to be recognized.
  • the second preset matching degree threshold is 0.6-0.9, and the first preset number of thresholds is 8000-11000.
  • the determining unit is further configured to: determine the shape feature, the texture feature set, and the color feature set as an initialization group; and the image recognition device further includes: an operation unit, configured to sequentially perform the initialization group The mutation, the intersection and the selection operation are performed to form an evolutionary group; the third determining unit is configured to determine whether the evolutionary group satisfies the second predetermined condition; and the determining unit is further configured to: when the evolutionary group satisfies the second predetermined condition, the evolutionary group is The corresponding image is determined as a first evolved image; the operation unit is further configured to: when the evolved group does not satisfy the second preset condition, use the evolved group as an initialization group, and continue to perform mutation, crossover, and selection operations, wherein the second pre- The condition that the number of the mutation, the intersection, and the selection operation is not less than the second preset number threshold and/or the maximum matching degree of the image corresponding to the evolution group and the image in the preset image library is not less than the third preset matching degree
  • the second preset number threshold is 8000-11000
  • the third preset matching threshold is 0.6-0.9.
  • the mutation, crossover, and selection operations can be implemented by a differential evolution algorithm.
  • the method further includes: an updating unit, configured to add the matching image to the first preset image library, and update the image feature weight table.
  • the association between the image feature weight table and the user's preference is realized, and the image to be identified is filtered according to the user's preference.
  • the realization of the image features further improves the accuracy of image recognition of the image to be recognized.
  • the technical solution of the third aspect of the present invention provides a computer device including a processor for implementing any one of the technical solutions of the first aspect of the present invention when the processor is configured to execute a computer program stored in the memory.
  • the steps of the image recognition method are described in detail below.
  • the computer device includes a processor, and the processor is configured to implement the steps of the image recognition method according to any one of the technical solutions of the first aspect of the present invention when the computer program stored in the memory is executed, thereby having the above All the beneficial effects of the image recognition method according to any one of the technical solutions of the first aspect of the present invention are not described herein.
  • the technical solution of the fourth aspect of the present invention provides a computer readable storage medium having stored thereon a computer program, which when executed by a processor, implements an image of any one of the technical solutions of the first aspect of the present invention The steps to identify the method.
  • the computer readable storage medium has stored thereon a computer program, and when the computer program is executed by the processor, implements the steps of the image recognition method according to any one of the technical solutions of the first aspect of the present invention, and thus has All the beneficial effects of the image recognition method according to any one of the technical solutions of the first aspect of the present invention are not described herein.
  • FIG. 1 shows a schematic flow chart of an image recognition method according to an embodiment of the present invention
  • FIG. 2 shows a schematic block diagram of an image recognition apparatus in accordance with one embodiment of the present invention
  • FIG. 3 shows a schematic flow chart of an image recognition method according to another embodiment of the present invention.
  • FIG. 1 shows a schematic flow chart of an image recognition method in accordance with one embodiment of the present invention.
  • an image recognition method includes: step S102, determining a shape feature of an image to be recognized; and step S104, extracting a texture feature and a color feature of the image to be identified; and step S106, according to the first Determining an image feature weight table in the image library, determining a weight of each texture feature of the image to be recognized and a weight of each color feature; and in step S108, filtering the texture feature and the color feature that are greater than the preset weight to form a texture respectively a feature set and a color feature set; step S110, determining a first evolved image corresponding to the image to be recognized according to the shape feature, the texture feature set, and the color feature set; and step S112, comparing the first evolved image with the second preset image library The matching degree of an image determines the matching image of the image to be identified.
  • the feature of the image to be recognized is extracted, by using the image feature weight table in the first preset image library. Determining the weight of each texture feature of the image to be identified and the weight of each color feature, and then filtering the texture features and color features that are greater than the preset weights to form the texture feature set and the color feature set respectively, further optimizing the waiting for Identifying the features extracted from the image is beneficial to improve the recognition accuracy of the image to be recognized.
  • the first evolution image corresponding to the image to be identified is determined according to the shape feature, the texture feature set and the color feature set, and the image to be recognized is further optimized.
  • the matching degree, the matching image of the image to be identified is determined, and the The recognition of the image to be recognized improves the accuracy of the image recognition, and the number of sample images in the second preset image library only needs to include a small number of sample images containing the main image features to realize the recognition of the image to be recognized, and the sample image is reduced. The number saves hardware resources and reduces the difficulty of machine training.
  • the area suggestion frame is divided by the area suggestion network, and the image in the area suggestion frame is the image to be recognized, and the image to be recognized includes the image of the food material and the background image of the position, and is other
  • the image of the object is occluded, and the method of the present invention can screen the image features of the food material, and then obtain the first evolution image, so that the image of the sample material in the second preset image library can be matched, and then the matching degree is compared. To achieve image recognition of the image to be recognized.
  • the preset weight is 0.4
  • the first preset image library is a user's preference library
  • the image feature weight table in the first preset image library is based on any image feature in the first preset image library in all image features.
  • the weight of the image to be identified is determined by the gray level co-occurrence matrix algorithm, and the color feature of the image to be recognized can be extracted by a histogram algorithm.
  • the size of the image to be identified is M ⁇ N
  • the gray level is L
  • G ⁇ 0, 1, 2, ..., L-1 ⁇
  • f(x, y) is the gray of the pixel at the coordinates (x, y).
  • Degree level a co-occurrence matrix of each image is an L ⁇ L matrix T[t ij ] L ⁇ L
  • the elements in T are the spatial relationship of the gray level of the image, Extracted texture features p ij ,
  • the number of values, nk is the number of pixels in the image with a pixel feature value of k, and N is the total number of image pixels.
  • determining a shape feature of the image to be recognized includes: extracting a corner feature of the image to be recognized; calculating a matching degree of the corner feature with any shape in the preset shape library to determine a maximum matching degree. Determining whether the maximum matching degree is greater than the first preset matching degree threshold; if it is determined that the maximum matching degree is greater than the first preset matching degree threshold, determining that the shape corresponding to the maximum matching degree is the shape feature of the image to be identified; if determining the maximum matching degree If it is not greater than the first preset matching degree threshold, the corner feature of the image to be recognized is continuously extracted.
  • the maximum matching degree is determined by extracting the corner feature of the image to be recognized, the matching degree of the corner feature with any shape in the preset shape library, and then the maximum matching degree is greater than the first preset matching.
  • the threshold is determined, the shape corresponding to the maximum matching degree is determined as the shape feature of the image to be recognized, the shape feature of the image to be recognized is determined, and the accuracy of the shape feature is determined, which is beneficial to further improve the accuracy of image recognition. .
  • the first preset matching degree threshold is 0.5-0.8, and the corner feature extraction of the image to be identified can be implemented by a Harris corner feature extraction algorithm.
  • the shape space is constructed according to the shape feature, and then the texture feature in the texture feature set and the color feature in the color feature set are randomly selected and sequentially set in the shape space to form a second evolution image, and the second evolution image is improved.
  • the diversity of the evolved image is advantageous for improving the efficiency of image recognition, and is also beneficial for reducing the number of sample images, by judging whether the second evolved image satisfies the first preset condition, and when the second evolved image satisfies the first preset condition, Determining that the second evolution image is the first evolution image, and continuing to form a new second evolution image when the second evolution image does not satisfy the first preset condition, further reducing the first evolution image and the second preset image library
  • the image matching difficulty increases the accuracy of the recognition of the image to be recognized.
  • the second preset matching degree threshold is 0.6-0.9, and the first preset number of thresholds is 8000-11000.
  • determining, according to the shape feature, the texture feature set, and the color feature set, the first evolution image corresponding to the image to be identified including: determining the shape feature, the texture feature set, and the color feature set to be initialized a group; mutating, intersecting, and selecting operations are sequentially performed on the initialization group to form an evolutionary group; determining whether the evolutionary group satisfies a second predetermined condition; if the evolutionary group is determined to satisfy the second predetermined condition, determining an image corresponding to the evolved group as The first evolutionary image; if it is determined that the evolutionary group does not satisfy the second predetermined condition, the evolutionary group is used as an initialization group, and the mutation, intersection, and selection operations are sequentially performed, wherein the second preset condition is mutation, intersection, and selection operation.
  • the number of times is not less than the second preset number threshold and/or the maximum matching degree of the image corresponding to the evolution group and the image in the preset image library is not less than the third preset matching degree threshold
  • the second preset number threshold is 8000-11000
  • the third preset matching threshold is 0.6-0.9.
  • the mutation, crossover, and selection operations can be implemented by a differential evolution algorithm.
  • the method further includes: adding the matching image to the first A preset image library and update the image feature weight table.
  • the association between the image feature weight table and the user's preference is realized, and the image to be identified is filtered according to the user's preference.
  • the realization of the image features further improves the accuracy of image recognition of the image to be recognized.
  • FIG. 2 shows a schematic block diagram of an image recognition device 200 in accordance with one embodiment of the present invention.
  • an image recognition apparatus 200 includes: a determining unit 204, configured to determine a shape feature of an image to be recognized; and an extracting unit, configured to extract a texture feature and a color feature of the image to be recognized;
  • the determining unit 204 is further configured to: determine, according to the image feature weight table in the first preset image library, the weight of each texture feature of the image to be identified and the weight of each color feature;
  • the screening unit 206 is configured to use the screening weight to be greater than Determining the texture feature and the color feature to form the texture feature set and the color feature set respectively;
  • the determining unit 204 is further configured to: determine the first evolution image corresponding to the image to be identified according to the shape feature, the texture feature set, and the color feature set;
  • the comparing unit 208 is configured to compare the matching degree of any one of the first evolution image and the second preset image library to determine a matching image of the image to be identified.
  • the feature of the image to be recognized is extracted, by using the image feature weight table in the first preset image library. Determining the weight of each texture feature of the image to be identified and the weight of each color feature, and then filtering the texture features and color features that are greater than the preset weights to form the texture feature set and the color feature set respectively, further optimizing the waiting for Identifying the features extracted from the image is beneficial to improve the recognition accuracy of the image to be recognized.
  • the first evolution image corresponding to the image to be identified is determined according to the shape feature, the texture feature set and the color feature set, and the image to be recognized is further optimized.
  • the matching degree, the matching image of the image to be identified is determined, and the The recognition of the image to be recognized improves the accuracy of the image recognition, and the number of sample images in the second preset image library only needs to include a small number of sample images containing the main image features to realize the recognition of the image to be recognized, and the sample image is reduced. The number saves hardware resources and reduces the difficulty of machine training.
  • the area suggestion frame is divided by the area suggestion network, and the image in the area suggestion frame is the image to be recognized, and the image to be recognized includes the image of the food material and the background image of the position, and is other
  • the image of the object is occluded, and the method of the present invention can screen the image features of the food material, and then obtain the first evolution image, so that the image of the sample material in the second preset image library can be matched, and then the matching degree is compared. To achieve image recognition of the image to be recognized.
  • the preset weight is 0.4
  • the first preset image library is a user's preference library
  • the image feature weight table in the first preset image library is based on any image feature in the first preset image library in all image features.
  • the weight of the image to be identified is determined by the gray level co-occurrence matrix algorithm, and the color feature of the image to be recognized can be extracted by a histogram algorithm.
  • the size of the image to be identified is M ⁇ N
  • the gray level is L
  • G ⁇ 0, 1, 2, ..., L-1 ⁇
  • f(x, y) is the gray of the pixel at the coordinates (x, y).
  • Degree level a co-occurrence matrix of each image is an L ⁇ L matrix T[t ij ] L ⁇ L
  • the elements in T are the spatial relationship of the gray level of the image, Extracted texture features p ij ,
  • the number of values, nk is the number of pixels in the image with a pixel feature value of k, and N is the total number of image pixels.
  • the extracting unit is further configured to: extract a corner feature of the image to be recognized;
  • the image recognition device 200 further includes: a calculating unit 210, configured to calculate any one of the corner feature and the preset shape library The matching degree of the shape to determine the maximum matching degree;
  • the first determining unit 212 is configured to determine whether the maximum matching degree is greater than the first preset matching degree threshold;
  • the determining unit 204 is further configured to: determine the maximum matching degree in the first determining unit 212 When the value is greater than the first preset matching degree threshold, the shape corresponding to the maximum matching degree is determined as the shape feature of the image to be identified;
  • the extracting unit is further configured to: determine, in the first determining unit 212, that the maximum matching degree is not greater than the first preset matching degree threshold.
  • the corner feature of the image to be recognized is continuously extracted.
  • the maximum matching degree is determined by extracting the corner feature of the image to be recognized, the matching degree of the corner feature with any shape in the preset shape library, and then the maximum matching degree is greater than the first preset matching.
  • the threshold is determined, the shape corresponding to the maximum matching degree is determined as the shape feature of the image to be recognized, the shape feature of the image to be recognized is determined, and the accuracy of the shape feature is determined, which is beneficial to further improve the accuracy of image recognition. .
  • the first preset matching degree threshold is 0.5-0.8, and the corner feature extraction of the image to be identified can be implemented by a Harris corner feature extraction algorithm.
  • the corner response value D is greater than the set threshold and is the local maximum in the neighborhood of the point, the point is extracted as a corner feature.
  • the method further includes: a construction unit 214, configured to construct a shape space according to the shape feature; and a selection unit 216, configured to randomly select the texture feature and the color feature set in the texture feature set
  • the color features are respectively disposed in the shape space to form a second evolution image
  • the second determining unit 218 is configured to determine whether the second evolution image satisfies the first preset condition
  • the determining unit 204 is further configured to: in the second determining unit 218: determining that the second evolution image is the first evolution image when the second evolution image meets the first preset condition
  • the selecting unit 216 is further configured to: determine, by the second determining unit 218, that the second evolution image does not satisfy the first preset condition And re-randomly selecting the texture features in the texture feature set and the color features in the color feature set are respectively set in the shape space to form a new second evolution image, wherein the first preset condition is the second evolution image and The maximum matching degree of the image in the second preset image library is not less
  • the shape space is constructed according to the shape feature, and then the texture feature in the texture feature set and the color feature in the color feature set are randomly selected and sequentially set in the shape space to form a second evolution image, and the second evolution image is improved.
  • the diversity of the evolved image is advantageous for improving the efficiency of image recognition, and is also beneficial for reducing the number of sample images, by judging whether the second evolved image satisfies the first preset condition, and when the second evolved image satisfies the first preset condition, Determining that the second evolution image is the first evolution image, and continuing to form a new second evolution image when the second evolution image does not satisfy the first preset condition, further reducing the first evolution image and the second preset image library
  • the image matching difficulty increases the accuracy of the recognition of the image to be recognized.
  • the second preset matching degree threshold is 0.6-0.9, and the first preset number of thresholds is 8000-11000.
  • the determining unit 204 is further configured to: determine the shape feature, the texture feature set, and the color feature set as an initialization group; the image recognition apparatus 200 further includes: an operation unit 220, configured to perform initialization The group sequentially performs mutation, crossover, and selection operations to form an evolutionary group.
  • the third determining unit 222 is configured to determine whether the evolved group satisfies the second preset condition.
  • the determining unit 204 is further configured to: meet the second preset condition in the evolved group.
  • the image corresponding to the evolutionary group is determined as the first evolution image; the operation unit 220 is further configured to: when the evolutionary group does not satisfy the second preset condition, use the evolutionary group as the initialization group, and continue to perform mutation, crossover, and selection operations in sequence.
  • the second preset condition is that the number of the mutation, the intersection, and the selection operation is not less than the second preset number threshold and/or the maximum matching degree between the image corresponding to the evolution group and the image in the preset image library is not less than the third pre-predetermined Set the match threshold.
  • the shape feature, the texture feature set, and the color feature set as the initialization group, and then performing multiple mutations, intersections, and selection operations, multiple evolutions are performed to obtain the first evolution image, which further reduces the
  • the matching difficulty of the images in the two preset image libraries improves the accuracy of image recognition and improves the efficiency of image recognition.
  • the second preset number threshold is 8000-11000
  • the third preset matching threshold is 0.6-0.9.
  • the mutation, crossover, and selection operations can be implemented by a differential evolution algorithm.
  • the method further includes: an updating unit 224, configured to add a matching image to the first preset image library, and update the image feature weight table.
  • the association between the image feature weight table and the user's preference is realized, and the image to be identified is filtered according to the user's preference.
  • the realization of the image features further improves the accuracy of image recognition of the image to be recognized.
  • the computer apparatus comprises a processor for performing the steps of the image recognition method according to any one of the embodiments of the present invention described above when executing the computer program stored in the memory.
  • the computer device comprises a processor for performing the steps of the image recognition method according to any of the above-mentioned embodiments of the present invention when executing the computer program stored in the memory, and thus having the above-described implementation of the present invention
  • a processor for performing the steps of the image recognition method according to any of the above-mentioned embodiments of the present invention when executing the computer program stored in the memory, and thus having the above-described implementation of the present invention.
  • a computer readable storage medium according to an embodiment of the present invention, wherein a computer program is stored thereon, and when the computer program is executed by the processor, the steps of the image recognition method according to any one of the embodiments of the present invention described above are implemented.
  • a computer readable storage medium having stored thereon a computer program executed by a processor to implement the steps of the image recognition method of any of the above-described embodiments of the present invention, and thus having the above-described present invention
  • the entire beneficial effects of the image recognition method of any of the embodiments proposed in the embodiments are not described herein.
  • FIG. 3 shows a schematic flow chart of an image recognition method according to another embodiment of the present invention.
  • step S312 the image recognition method according to another embodiment of the present invention, after inputting the food material image in step S302, performing step S304 area suggestion box division, step S306, selecting a certain area suggestion box, step S308, extracting Corner feature, then proceeding to step S312 to match the shape trend in the shape library, step S314, determining the shape feature of the food item, step S318, adding to the initialization group, and performing step S310 in the selected certain area suggestion box, Extracting the color feature and the texture feature, step S318, adding to the initialization group, and then performing step S320 according to the user ingredient preference library, initializing the color feature weight and step S324, initializing the texture feature weight, and performing step S322 after step S320 to filter out the weight is greater than a color feature set of 0.4, after step S324, step S326 is performed to filter out a texture feature set having a weight greater than 0.4, in which the step S328 randomly selects the color feature and the step S330 randomly selects the texture feature, and then proceeds to step S3
  • Image library, evaluating evolution results, step S334, according to The food image library calculates the matching degree, and proceeds to step S336 to determine whether the evaluation condition is satisfied. If the determination is yes, the process proceeds to step S338 to form a first evolution image. If the evaluation condition is not satisfied, the process proceeds to step S328 to re-randomly select the color feature and reselect. Texture features.
  • the present invention provides an image recognition method, apparatus, computer device and readable storage medium, according to a shape feature, a texture feature set and a color feature set according to an image to be recognized. Determining a first evolution image corresponding to the image to be identified, and then comparing a matching degree of the image of the first evolution image with the second preset image library, determining a matching image of the image to be recognized, improving the accuracy of image recognition, and reducing The number of sample images saves hardware resources.
  • the units in the apparatus of the present invention can be combined, divided, and deleted according to actual needs.
  • ROM Read-Only Memory
  • RAM Random Access Memory
  • PROM Programmable Read-Only Memory
  • EPROM Erasable Programmable Read Only Memory
  • OTPROM One-Time Programmable Read-Only Memory
  • EEPROM Electronically-Erasable Programmable Read-Only Memory
  • CD-ROM Compact Disc Read-Only Memory

Abstract

L'invention concerne un procédé de reconnaissance d'image, un appareil, un dispositif informatique et un support d'informations lisible, le procédé de reconnaissance d'image consistant : à déterminer une caractéristique de forme d'une image à reconnaître (S102) ; à extraire des caractéristiques de texture et des caractéristiques de couleur de l'image à reconnaître (S104) ; selon une table de poids caractéristiques d'images dans une première bibliothèque prédéfinie d'images, à déterminer les poids de chaque caractéristique de texture et les poids de chaque caractéristique de couleur de l'image à reconnaître (S106) ; à cribler des caractéristiques de texture et des caractéristiques de couleur dont les poids sont supérieurs à un poids prédéfini, de façon à former un ensemble de caractéristiques de texture et un ensemble de caractéristiques de couleur respectivement (S108) ; en fonction de la caractéristique de forme, de l'ensemble de caractéristiques de texture et de l'ensemble de caractéristiques de couleur, à déterminer une première image d'évolution correspondant à l'image à reconnaître (S110) ; à comparer un degré de correspondance entre la première image d'évolution et toute image dans une seconde bibliothèque prédéfinie d'images, et déterminer une image correspondante pour l'image à reconnaître (S112). Le présent procédé augmente la précision de la reconnaissance d'image, réduit le nombre d'images d'échantillon et économise des ressources matérielles.
PCT/CN2018/090970 2017-08-04 2018-06-13 Procédé de reconnaissance d'image, appareil, dispositif informatique et support d'informations lisible WO2019024610A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710661653.8A CN107480711B (zh) 2017-08-04 2017-08-04 图像识别方法、装置、计算机设备和可读存储介质
CN201710661653.8 2017-08-04

Publications (1)

Publication Number Publication Date
WO2019024610A1 true WO2019024610A1 (fr) 2019-02-07

Family

ID=60597512

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/090970 WO2019024610A1 (fr) 2017-08-04 2018-06-13 Procédé de reconnaissance d'image, appareil, dispositif informatique et support d'informations lisible

Country Status (2)

Country Link
CN (1) CN107480711B (fr)
WO (1) WO2019024610A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113642552A (zh) * 2020-04-27 2021-11-12 上海高德威智能交通系统有限公司 一种图像中目标对象的识别方法、装置、系统及电子设备

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107480711B (zh) * 2017-08-04 2020-09-01 合肥美的智能科技有限公司 图像识别方法、装置、计算机设备和可读存储介质
CN109993178B (zh) * 2017-12-29 2024-02-02 华为技术有限公司 一种特征数据生成和特征匹配方法及装置
CN108224894B (zh) * 2018-01-08 2020-04-28 合肥美的智能科技有限公司 基于深度学习的食材新鲜度识别方法、装置、冰箱和介质
CN109030480B (zh) * 2018-08-16 2021-03-19 湖南友哲科技有限公司 样品分析方法、装置、可读存储介质及计算机设备
CN111353333B (zh) * 2018-12-21 2023-10-20 九阳股份有限公司 一种食材识别方法、家电设备及食材识别系统
CN110222789B (zh) * 2019-06-14 2023-05-26 腾讯科技（深圳）有限公司 图像识别方法及存储介质
CN111476289B (zh) * 2020-04-03 2024-04-19 江苏提米智能科技有限公司 一种基于特征库的鱼群识别方法、装置、设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101211341A (zh) * 2006-12-29 2008-07-02 上海芯盛电子科技有限公司 图像智能模式识别搜索方法
US20090289942A1 (en) * 2008-05-20 2009-11-26 Timothee Bailloeul Image learning, automatic annotation, retrieval method, and device
CN101635835A (zh) * 2008-07-25 2010-01-27 深圳市信义科技有限公司 智能视频监控方法及系统
CN104537376A (zh) * 2014-11-25 2015-04-22 深圳创维数字技术有限公司 一种识别台标的方法及相关设备、系统
CN107480711A (zh) * 2017-08-04 2017-12-15 合肥美的智能科技有限公司 图像识别方法、装置、计算机设备和可读存储介质

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090116757A1 (en) * 2007-11-06 2009-05-07 Copanion, Inc. Systems and methods for classifying electronic documents by extracting and recognizing text and image features indicative of document categories
CN104298988B (zh) * 2014-08-21 2017-08-25 华南理工大学 一种基于视频图像局部特征匹配的财物保护方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101211341A (zh) * 2006-12-29 2008-07-02 上海芯盛电子科技有限公司 图像智能模式识别搜索方法
US20090289942A1 (en) * 2008-05-20 2009-11-26 Timothee Bailloeul Image learning, automatic annotation, retrieval method, and device
CN101635835A (zh) * 2008-07-25 2010-01-27 深圳市信义科技有限公司 智能视频监控方法及系统
CN104537376A (zh) * 2014-11-25 2015-04-22 深圳创维数字技术有限公司 一种识别台标的方法及相关设备、系统
CN107480711A (zh) * 2017-08-04 2017-12-15 合肥美的智能科技有限公司 图像识别方法、装置、计算机设备和可读存储介质

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113642552A (zh) * 2020-04-27 2021-11-12 上海高德威智能交通系统有限公司 一种图像中目标对象的识别方法、装置、系统及电子设备
CN113642552B (zh) * 2020-04-27 2024-03-08 上海高德威智能交通系统有限公司 一种图像中目标对象的识别方法、装置、系统及电子设备

Also Published As

Publication number Publication date
CN107480711B (zh) 2020-09-01
CN107480711A (zh) 2017-12-15

Similar Documents

Publication Publication Date Title
WO2019024610A1 (fr) Procédé de reconnaissance d'image, appareil, dispositif informatique et support d'informations lisible
CN110163234B (zh) 一种模型训练方法、装置和存储介质
CN109409398A (zh) 图像处理装置、图像处理方法以及存储介质
CN115017418B (zh) 基于强化学习的遥感影像推荐系统及方法
WO2019134531A1 (fr) Procédé destiné à trouver une ligne de couture optimale d'image panoramique
JP2013125322A (ja) 学習装置、プログラム及び学習方法
US11373309B2 (en) Image analysis in pathology
US20220414892A1 (en) High-precision semi-automatic image data labeling method, electronic apparatus, and storage medium
CN113836373B (zh) 一种基于密度聚类的投标信息处理方法、设备及存储介质
CN115830335A (zh) 一种基于自适应阈值算法的orb图像特征提取方法
CN116012721B (zh) 一种基于深度学习的水稻叶片病斑检测方法
CN110910401A (zh) 半自动化图像分割数据标注方法、电子装置及存储介质
CN115937065A (zh) 显示模组的异物检测方法、装置、设备及存储介质
JP2020160543A (ja) 情報処理システムおよび情報処理方法
CN111612099B (zh) 基于局部排序差值细化模式的纹理图像分类方法及系统
CN108876776A (zh) 一种分类模型生成方法、眼底图像分类方法及装置
CN114707014B (zh) 一种基于fov的影像数据融合索引方法
CN112734747A (zh) 一种目标检测方法、装置、电子设备和存储介质
da Silva Teixeira et al. Reconstruction of frescoes by sequential layers of feature extraction
JP2011150626A (ja) 画像分類方法、装置、及びプログラム
CN114972702A (zh) 一种工业图像目标检测图像训练集的采样方法和存储介质
CN114418898A (zh) 一种基于目标重叠度计算和自适应调整的数据增强方法
WO2019127075A1 (fr) Procédé d'identification de l'année d'une pièce de monnaie, dispositif de terminal et support d'enregistrement lisible par ordinateur
CN114359090A (zh) 一种口腔ct影像的数据增强方法
CN114359300A (zh) 一种图像分割模型的优化方法、装置、系统及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18840184

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18840184

Country of ref document: EP

Kind code of ref document: A1