COLOR RECOGNITION THROUGH LEARNED COLOR CLUSTERS

• US 20180260974A1
• Filed: 03/09/2017
• Published: 09/13/2018
• Est. Priority Date: 03/09/2017
• Status: Active Grant

• Current Assignee: Micro Focus LLC (Open Text Corporation)
• Original Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company)
• Inventors: Cameron, Pashmina, Woods, Timothy
• Granted Patent: US 10,423,855 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/2321 using statistics or function optimisation, e.g. modelling of probability density functions
• CPC Class Codes: G06V 10/56 relating to colour
• CPC Class Codes: G06V 10/763 Non-hierarchical techniques, e.g. based on statistics of modelling distributions
• CPC Class Codes: G06V 2201/08 Detecting or categorising vehicles

Abstract

In some examples, a system includes a color cluster learning engine and a color recognition engine. The color cluster learning engine may be configured to obtain a set of training images, process the training images to obtain clusters of pixel colors for the training images, identify learned color clusters from the clusters of pixel colors obtained from the training images, and label the learned color clusters with color indicators. The color recognition engine may be configured to receive an input image for color identification, process the input image to obtain a particular cluster of pixel colors that covers the highest number of pixels in the input image, match the particular cluster to a particular learned color cluster labeled with a particular color indicator, and identify a color of the input image as specified by the particular color indicator.

Claims

1. A system comprising:

a color cluster learning engine configured to;

obtain a set of training images;

process the training images to obtain clusters of pixel colors for the training images;

identify learned color clusters from the clusters of pixel colors obtained from the training images; and

label the learned color clusters with color indicators; and

a color recognition engine configured to;

receive an input image for color identification;

process the input image to obtain a particular cluster of pixel colors that covers the highest number of pixels in the input image;

match the particular cluster to a particular learned color cluster labeled with a particular color indicator; and

identify a color of the input image as specified by the particular color indicator.

2. The system of claim 1, wherein the training images are represented in a color space including a luminance component;

and

wherein the color cluster learning engine is configured to process the training set to obtain clusters of pixel colors by applying a clustering process using a weighted distance metric that reduces a weight of the luminance component of a pixel color at ends of a luminance spectrum.

3. The system of claim 2, wherein the color recognition engine is configured to process the input image also using the weighted distance metric to obtain the clusters of pixel colors from the input image.

4. The system of claim 1, wherein:

the color cluster learning engine is further configured to represent each learned color cluster as the pixel color of a cluster center of the learned color cluster; and

the color recognition engine is configured to match the particular cluster obtained from the input image to the particular learned color cluster by determining that a center of the particular cluster has a lesser distance to the cluster center of the particular learned color cluster than to other cluster centers of other learned color clusters.

5. The system of claim 4, wherein the training images and the input images are represented in a color space including a luminance component and

wherein the color recognition engine is configured to compare distances between cluster centers using a weighted distance metric that reduces the luminance component of a pixel color at ends of a luminance spectrum.

6. The system of claim 1, wherein the color recognition engine is further configured assign a particular pixel of the input image to a cluster as part of processing the input image by:

calculating distances between the particular pixel and cluster centers of randomly initialized clusters;

calculating distances between the particular pixel and cluster centers of the learned color clusters;

determining biased distances by reducing the calculated distances to the cluster centers of the learned color clusters by a bias value; and

assigning the particular pixel to a cluster with a least calculated distance among the calculated distances to the cluster centers of the randomly initialized clusters and the biased distances.

7. The system of claim 1, wherein the color cluster learning engine is to label multiple learned color clusters with the same color indicator.

8. A method comprising:

accessing an input image for color recognition;

applying a clustering process to obtain clusters of pixel colors for the input image, the pixel colors represented in a color space including a luminance component, and wherein applying the clustering process comprises using a weighted distance metric that reduces the luminance component of a pixel color at ends of a luminance spectrum;

identifying a particular cluster among the obtained clusters that covers a highest number of pixels in the input image;

matching the particular cluster to a particular learned color cluster labeled with a particular color indicator; and

identifying a color for the input image as specified by the particular color indicator.

9. The method of claim 8, wherein matching comprises determining that the particular cluster has a lesser distance in the color space to a cluster center of the particular learned color cluster than to other cluster centers of other learned color clusters.

10. The method of claim 9, further comprising determining distances between the particular color clusters and cluster centers of learned color clusters using a weighted distance metric that reduces the luminance component of a pixel color at ends of a luminance spectrum.

11. The method of claim 10, wherein:

the weighted distance metric weights the luminance component of a pixel color with a first weight value at the ends of the luminance spectrum and a second weight value at a midpoint of the luminance spectrum; and

the second weight value is greater than the first weight value.

12. The method of claim 11, wherein weighted distance metric further weights the luminance component linearly between the ends of the luminance spectrum and the midpoint of the luminance spectrum.

13. The method of claim 11, wherein weighted distance metric further weights the luminance component between the ends of the luminance spectrum and the midpoint of the luminance spectrum according to an entropy curve.

14. The method of claim 8, further comprising, prior to accessing the input image:

obtaining a set of training images;

processing the training images to obtain clusters of pixel colors for the training images, the pixel colors represented in the color space including a luminance component;

identifying a set of learned color clusters that includes the particular learned color cluster from the training images; and

labeling the learned color clusters with color indicators including labeling the particular learned color cluster with the particular color indicator.

15. A non-transitory machine-readable medium comprising instructions executable by a processing resource to:

obtain a training set of vehicle images;

process the training set of vehicle images to obtain clusters of pixel colors for the training set of vehicle images, the pixel colors represented in a color space including a luminance component;

identify a set of learned color clusters from the training set of vehicle images;

correlate the learned color clusters to vehicle colors; and

receive an input vehicle image to extract a vehicle color from;

process the input vehicle image to obtain a particular cluster of pixel colors that covers the highest number of pixels in the input vehicle image, the pixel colors represented in the color space with the luminance component;

map the particular cluster to a particular learned color cluster associated with a particular vehicle color; and

identify a vehicle color of the input vehicle image as the particular vehicle color.

16. The non-transitory machine-readable medium of claim 15, wherein the instructions are executable by the processing resource to process the training set of vehicle images to obtain clusters of pixel colors by applying a clustering process using a weighted distance metric that reduces the luminance component of a pixel color at ends of a luminance spectrum.

17. The non-transitory machine-readable medium of claim 16, wherein the instructions are executable by the processing resource to process the input vehicle image also using the weighted distance metric to obtain the clusters of pixel colors from the input vehicle image.

18. The non-transitory machine-readable medium of claim 16, wherein:

the weighted distance metric weights the luminance component of a pixel color with a first weight value at the ends of the luminance spectrum and a second weight value at a midpoint of the luminance spectrum; and

the second weight value is greater than the first weight value.

19. The non-transitory machine-readable medium of claim 15, wherein the instructions are executable by the processing resource to map multiple learned color clusters to the particular vehicle color.

20. The non-transitory machine-readable medium of claim 19, wherein the multiple learned color clusters associated with the particular vehicle color include:

a first learned color cluster corresponding to the particular vehicle color at a first illumination level; and

a second learned color cluster corresponding to the particular vehicle color at a second illumination level different from the first illumination level.