Best image crop selection

Patent metadata

Field	Value
Publication number	US-10671887-B2
Application number	US-201715831638-A
Country	US
Kind code	B2
Filing date	Dec 5, 2017
Priority date	Dec 5, 2017
Publication date	Jun 2, 2020
Grant date	Jun 2, 2020

Abstract

Methods and apparatus, including computer program products, for creating a quality annotated training data set of images for training a quality estimating neural network. A set of images depicting a same object is received. The images in the set of images have varying image quality. A probe image whose quality is to be estimated is selected from the set of images. A gallery of images is selected from the set of images. The gallery of images does not include the probe image. The probe image is compared to each image in the gallery and a match score is generated for each image comparison. Based on the match scores, a quality value is determined for the probe image. The probe image and its associated quality value are added to a quality annotated training data set for the neural network.

First claim

What is claimed is:

1. A method for creating a quality annotated training data set of images for training a quality estimating neural network, comprising:
receiving a set of images depicting a same object, wherein the images in the set of images have varying image quality;
selecting, from the set of images, a probe image whose quality is to be estimated;
selecting, from the set of images, a gallery of images, wherein the gallery of images does not include the probe image;
comparing the probe image to each image in the gallery and generating a match score for each image comparison;
determining, based on the match scores, a quality value for the probe image; and
adding the probe image and its associated quality value to a quality annotated training data set for the neural network.

2. The method of claim 1 , further comprising:
repeating the selecting, selecting, comparing, determining and adding steps for each image in the set of images.

3. The method of claim 1 , wherein the object is one of: a person, a vehicle, and a face.

4. The method of claim 1 , wherein comparing the probe image to each image in the gallery of images includes:
determining a distance between a first feature vector representing the probe image and a second feature vector representing a gallery image; and
assigning a match score based on the distance between the first feature vector and the second feature vector.

5. The method of claim 4 , wherein a short distance indicates a high degree of identity similarity between the probe image and the gallery image.

6. The method of claim 1 , wherein the quality value for the probe image is determined as a function of a best match score for the probe image and the range of match scores for the probe image.

7. A system for creating a quality annotated training data set of images for training a quality estimating neural network, comprising:
a memory; and a processor,
wherein the memory contains instructions that when executed by the processor causes the processor to perform a method that includes:
receiving a set of images depicting a same object, wherein the images in the set of images have varying image quality;
selecting, from the set of images, a probe image whose quality is to be estimated;
selecting, from the set of images, a gallery of images, wherein the gallery of images does not include the probe image;
comparing the probe image to each image in the gallery and generating a match score for each image comparison;
determining, based on the match scores, a quality value for the probe image; and
adding the probe image and its associated quality value to a quality annotated training data set for the neural network.

8. The system of claim 7 , wherein the method further comprises:
repeating the selecting, selecting, comparing, determining and adding steps for each image in the set of images.

9. The system of claim 7 , wherein the object is one of: a person, a vehicle, and a face.

10. The system of claim 7 , wherein comparing the probe image to each image in the gallery of images includes:
determining a distance between a first feature vector representing the probe image and a second feature vector representing a gallery image; and
assigning a match score based on the distance between the first feature vector and the second feature vector.

11. The system of claim 10 , wherein a short distance indicates a high degree of identity similarity between the probe image and the gallery image.

12. The system of claim 7 , wherein the quality value for the probe image is determined as a function of a best match score for the probe image and the range of match scores for the probe image.

13. A computer program product for creating a quality annotated training data set of images for training a quality estimating neural network, comprising a computer readable storage medium having program instructions embodied therewith, wherein the computer readable storage medium is not a transitory signal per se, the program instructions being executable by a processor to perform a method comprising:
receiving a set of images depicting a same object, wherein the images in the set of images have varying image quality;
selecting, from the set of images, a probe image whose quality is to be estimated;
selecting, from the set of images, a gallery of images, wherein the gallery of images does not include the probe image;
comparing the probe image to each image in the gallery and generating a match score for each image comparison;
determining, based on the match scores, a quality value for the probe image; and
adding the probe image and its associated quality value to a quality annotated training data set for the neural network.

14. The computer program product of claim 13 , wherein the method further comprises:
repeating the selecting, selecting, comparing, determining and adding steps for each image in the set of images.

15. The computer program product of claim 13 , wherein comparing the probe image to each image in the gallery of images includes:
determining a distance between a first feature vector representing the probe image and a second feature vector representing a gallery image; and
assigning a match score based on the distance between the first feature vector and the second feature vector.

Assignees

• Axis Ab

Inventors

• Danielsson Niclas
• Skans Markus

Classifications

• G06K9/00771

  Physics · mapped topic

• G06K9/66

  Physics · mapped topic

• G06K9/4628

  Physics · mapped topic

• G06K9/6202

  Physics · mapped topic

• G06K9/6256 Primary

  Physics · mapped topic