Reducing scale estimate errors in shelf images
View Patent ↗Example image processing methods, apparatus/systems and articles of manufacture are disclosed herein. An example apparatus includes an image recognition application to identify matches between stored patterns and objects detected in a shelf image, where the shelf image has a shelf image scale estimate. The example apparatus further includes a scale corrector to calculate deviation values between sizes of (A) a first set of the objects detected in the shelf image and (B) a first set of the stored patterns matched with the first set of the objects and reduce an error of the shelf image scale estimate by calculating a scale correction value for the shelf image scale estimate based on the deviation values.
1. An image processing system comprising:
a pattern scale optimizer to determine optimized scale values for a first set of patterns of a plurality of patterns stored in a database, the pattern scale optimizer to determine the optimized scale values based on scale estimates associated with the first set of patterns;
a rescaler to generate rescaled versions of the first set of patterns using the optimized scale values for the first set of patterns and based on a shelf image scale estimate of a shelf image; and
an image recognition application to use the rescaled versions of the first set of patterns to identify matches between the first set of patterns and objects appearing in the shelf image.
2. The system of claim 1 , wherein, prior to determining the optimized scale values, the pattern scale optimizer is to determine that each of the first set of patterns has a number of scale estimates that satisfy a threshold.
3. The system of claim 2 , wherein the plurality of patterns includes a second set of patterns having a number of scale estimates that do not satisfy the threshold.
4. The system of claim 1 , wherein the optimized scale values are a median of the scale estimates associated with respective ones of the first set of patterns.
5. The system of claim 1 , further including a scale corrector to determine deviation values between sizes of the rescaled versions of the first set of patterns and the objects matched with the first set of patterns.
6. The system of claim 5 , wherein the scale corrector is to determine a scale correction value for the shelf image scale estimate based on the deviation values, the scale correction value to be used by the image recognition application during a subsequent analysis of the shelf image.
7. The system of claim 6 , wherein the scale corrector is to determine a new scale estimate for a first pattern of the first set of patterns based on a size of an object matched with a rescaled version of the first pattern and the scale correction value.
8. A method comprising:
analyzing, by executing an instruction with at least one processor, a shelf image a first time to identify matches between patterns stored in a database and objects appearing in the shelf image based on a first shelf image scale estimate of the shelf image;
determining, by executing an instruction with the at least one processor, a second shelf image scale estimate for the shelf image based on deviation values between (a) sizes of a first set of objects detected in the shelf image and (b) a first set of the patterns matched with the first set of objects; and
analyzing, by executing an instruction with the at least one processor, the shelf image a second time to identify matches between the patterns stored in the database and the objects appearing in the shelf image based on the second shelf image scale estimate.
9. The method of claim 8 , further including determining the deviation values by determining a difference in pixel dimension of a first scaled version of a first pattern of the first set of the patterns and a first object of the first set of objects match with the first pattern.
10. The method of claim 9 , further including, prior to analyzing the shelf image the first time, generating the first scaled version of the first pattern by rescaling the first pattern based on the first shelf image scale estimate.
11. The method of claim 8 , further including, prior to analyzing the shelf image the first time, rescaling the patterns to the first shelf image scale estimate.
12. The method of claim 8 , wherein the first set of the patterns have optimized scale values, further including, prior to analyzing the shelf image the first time, determining the optimized scale values for the first set of the patterns.
13. The method of claim 12 , wherein the optimized scale values for the first set of the patterns are based on arrays of scale estimates for respective ones of the first set of the patterns.
14. The method of claim 13 , wherein the first set of the patterns have a number of scale estimates that satisfy a threshold, and a second set of the patterns have a number of scale estimates that do not satisfy the threshold.
15. A tangible machine readable storage medium comprising instructions that, when executed, cause a machine to at least:
analyze a shelf image a first time using an image recognition technique to identify matches between stored patterns and objects detected in the shelf image, the shelf image having a shelf image scale estimate;
determine a scale correction value for the shelf image scale estimate based on deviations between sizes of a first set of objects detected in the shelf image and a first set of the patterns matched with the first set of objects, the first set of the patterns including a first pattern, the first pattern having an array of pattern scale estimates; and
determine a new scale estimate for the array of pattern scale estimates of the first pattern based on a size of an object matched with the first pattern and the scale correction value.
16. The tangible machine readable storage medium of claim 15 , wherein the instructions, when executed, cause the machine to, prior to analyzing the shelf image the first time:
determine a first optimized scale value for the first pattern based on the array of pattern scale estimates; and
generate a first scaled version of the first pattern based on the first optimized scale value and the shelf image scale estimate, the first scaled version of the first pattern to be used by the image recognition technique during the first time the shelf image is analyzed.
17. The tangible machine readable storage medium of claim 16 , wherein the instructions, when executed, cause the machine to, subsequent to determining the new scale estimate, determine a second optimized scale value for the first pattern based on the array of pattern scale estimates including the new scale estimate.
18. The tangible machine readable storage medium of claim 17 , wherein the instructions, when executed, cause the machine to analyze the shelf image a second time using the image recognition technique based on the second optimized scale value for the first pattern.
19. The tangible machine readable storage medium of claim 18 , wherein the shelf image scale estimate is a first shelf image scale estimate, and wherein the instructions, when executed, cause the machine to determine a second shelf image scale estimate based on the scale correction value.
20. The tangible machine readable storage medium of claim 19 , wherein the instructions, when executed, cause the machine to, prior to analyzing the shelf image the second time, generate a second rescaled version of the first pattern based on the second optimized scale value and the second shelf image scale estimate.