Methods and apparatus to perform image analyses in a computing environment
View Patent ↗An apparatus includes a feature extractor to generate image descriptors based on retail product tag images corresponding to a retailer category; a probability density function generator to generate a probability density function of probability values corresponding to visual features represented in the image descriptors; a sample selector to select ones of the probability values based on a sample selection algorithm that identifies positions in the probability density function of the ones of the probability values to be selected; a category signature generator to generate a category signature based on the selected ones of the probability values; and a processor to train a convolutional neural network (CNN) based on a feature descriptor and one of the retail product tag images, the feature descriptor including the category signature concatenated to one of the image descriptors, the training to cause the CNN to classify the one of the retail product tag images as a type of product tag.
1. An apparatus comprising:
a feature extractor to generate a plurality of image descriptors based on a plurality of retail product tag images corresponding to a retailer category, the plurality of image descriptors including values representative of one or more visual features of the plurality of retail product tag images;
a probability density function generator to generate a probability density function of probability values corresponding to the visual features represented in the plurality of image descriptors of the retailer category;
a sample selector to: generate position values based on a sample selection algorithm seeded by a retailer seed value corresponding to the retailer category; and
select ones of the probability values in the probability density function at positions identified by the position values;
a category signature generator to generate a category signature based on the selected ones of the probability values; and
a processor to train a convolutional neural network based on a feature descriptor and at least one of the retail product tag images, the feature descriptor including the category signature concatenated to one of the image descriptors, a bit length of the feature descriptor including a first number of bits of the one of the image descriptors and a second number of bits of the category signature, the training of the convolutional neural network to cause the convolutional neural network to classify the at least one of the retail product tag images as one of a plurality of types of product tags.
2. The apparatus as defined in claim 1 , wherein the probability density function generator is to model the probability density function of the probability values with a Gaussian distribution, the Gaussian distribution to approximate a real distribution.
3. The apparatus as defined in claim 1 , further including a seed generator to generate the retailer seed value based on a retailer identifier of the retailer category.
4. The apparatus as defined in claim 3 , wherein the seed generator is implemented by a random-number generator to generate the retailer seed value as a unique random seed.
5. The apparatus as defined in claim 1 , wherein the category signature is a length representing a number of samples selected by the sample selector, the samples corresponding to ones of the probability values.
6. The apparatus as defined in claim 1 , wherein the category signature is reproduceable based on the same retailer seed value.
7. The apparatus as defined in claim 1 , further including a memory to store the category signature in association with a retailer identifier corresponding to the retailer category, the category signature different from a second category signature generated by the category signature generator for a second retailer category.
8. A non-transitory computer readable storage medium comprising instructions that, when executed, cause a processor to at least:
generate a plurality of image descriptors based on a plurality of retail product tag images corresponding to a retailer category, the plurality of image descriptors including values representative of one or more visual features of the plurality of retail product tag images;
generate a probability density function of probability values corresponding to the visual features represented in the plurality of image descriptors of the retailer category;
generate position values based on a sample selection algorithm seeded by a retailer seed value corresponding to the retailer category, select ones of the probability values in the probability density function at positions identified by the position values;
generate a category signature based on the selected ones of the probability values; and
train a convolutional neural network based on a feature descriptor and at least one of the retail product tag images, the feature descriptor including the category signature concatenated to one of the image descriptors, a bit length of the feature descriptor including a first number of bits of the one of the image descriptors and a second number of bits of the category signature, the training of the convolutional neural network to cause the convolutional neural network to classify the at least one of the retail product tag images as one of a plurality of types of product tags.
9. The non-transitory computer readable storage medium as defined in claim 8 , wherein the instructions, when executed, cause the processor to model the probability density function of the probability values with a Gaussian distribution, the Gaussian distribution to approximate a real distribution.
10. The non-transitory computer readable storage medium as defined in claim 8 , wherein the instructions, when executed, cause the processor to generate the retailer seed value based on a retailer identifier of the retailer category.
11. The non-transitory computer readable storage medium as defined in claim 10 , wherein the instructions, when executed, cause the processor to implement a random-number generator to generate the retailer seed value as a unique random seed.
12. The non-transitory computer readable storage medium as defined in claim 10 , wherein the instructions, when executed, cause the processor to generate the category signature as a length representing a number of samples corresponding to ones of the probability values.
13. The non-transitory computer readable storage medium as defined in claim 8 , wherein the category signature is reproduceable based on the same retailer seed value.
14. The non-transitory computer readable storage medium as defined in claim 8 , wherein the instructions, when executed, cause the processor to store the category signature in association with a retailer identifier corresponding to the retailer category, the category signature different from a second category signature generated for a second retailer category.
15. A method comprising:
generating, by executing an instruction with a processor, a plurality of image descriptors based on a plurality of retail product tag images corresponding to a retailer category, the plurality of image descriptors including values representative of one or more visual features of the plurality of retail product tag images;
generating, by executing an instruction with the processor, a probability density function of probability values corresponding to the visual features represented in the plurality of image descriptors of the retailer category;
generating, by executing an instruction with the processor, position values based on a sample selection algorithm seeded by a retailer seed value corresponding to the retailer category, selecting, by executing an instruction with the processor, ones of the probability values identifying positions in the probability density function at positions identified by the position values;
generating, by executing an instruction with the processor, a category signature based on the selected ones of the probability values; and
training, by executing an instruction with the processor, a convolutional neural network based on a feature descriptor and at least one of the retail product tag images, the feature descriptor including the category signature concatenated to one of the image descriptors, a bit length of the feature descriptor including a first number of bits of the one of the image descriptors and a second number of bits of the category signature, the training of the convolutional neural network to cause the convolutional neural network to classify the at least one of the retail product tag images as one of a plurality of types of product tags.
16. The method as defined in claim 15 , further including modeling the probability density function with a Gaussian distribution to approximate a real distribution.
17. The method as defined in claim 15 , further including generating a retailer seed value based on a retailer identifier of the retailer category.
18. The method as defined in claim 17 , further including implementing a random-number generator to generate the retailer seed value as a unique random seed.
19. The method as defined in claim 15 , wherein the category signature is a length representing a number of samples corresponding to ones of the probability values.
20. The method as defined in claim 15 , wherein the category signature is reproduceable based on the same retailer seed value.
21. The method as defined in claim 15 , further including storing the category signature in association with a retailer identifier corresponding to the retailer category, the category signature different from a second category signature generated by a category signature generator for a second retailer category.