Logo detection system for automatic image search engines
A system and method that distinguishes between logos and other categories of images, such as natural images, cartoon images, and computer-processed or generated images (“concept images”). The system receives a query image, which may, for example, be intended to be evaluated by an image search and comparison engine to identify matches to a catalog of images. The system evaluates characteristics of the query image, such as the gray-levels in the image, the edge crossings in the image, and the gradient magnitudes in the image, to identify whether the image is a logo. Based on identifying the image is a logo, the logo query image may be excluded from being evaluated by the image search and comparison engine.
1. A method in a computing system for determining whether a query image is a logo and preventing identified logos from being searched within a catalog of images, the method comprising:
receiving a query image that is to be searched against a catalog of images, wherein the query image is comprised of a plurality of pixels;
evaluating pixel-values associated with each of the plurality of pixels of the query image to calculate a first factor indicative of whether the query image is a natural image;
evaluating a plurality of edge crossings in the query image to calculate a second factor indicative of whether the query image is a cartoon image;
evaluating gradient magnitudes of the query image to calculate a third factor indicative of whether the query image is a concept image, wherein evaluating the gradient magnitudes comprises evaluating whether a distribution of the gradient magnitudes resembles a power function having a negative exponent;
associating a weighting with each of the first, second, and third factors;
determining whether the query image is a logo based on the weighted first, second, and third factors;
using the query image to search within the catalog of images if the query image is not a logo; and
not using the query image to search within the catalog of images if the query image is a logo.
2. The method of claim 1 , further comprising, after receiving the query image, processing the query image by:
converting the query image to grayscale, and
resizing the query image.
3. The method of claim 1 , wherein evaluating the pixel-values associated with each of the plurality of pixels of the query image comprises:
calculating the number of pixels in the query image having each of the pixel-values; and
calculating an entropy of the number of pixels in the query image having each of the pixel-values.
4. The method of claim 1 , wherein evaluating the plurality of edge crossings in the query image comprises:
calculating the number of edge crossings in the query image; and
calculating an entropy of the number of edge crossings in the query image.
5. The method of claim 1 , wherein evaluating the gradient magnitudes of the query image further comprises:
calculating the gradient magnitudes of the query image; and
generating the distribution of the gradient magnitudes in the query image.
6. The method of claim 1 , wherein the evaluation of whether the distribution of gradient magnitudes resembles the power function having a negative exponent is based on calculating a linear approximation of the logarithm of the distribution and comparing the calculated linear approximation with the logarithm of the distribution.
7. The method of claim 1 , wherein the query image is excerpted from a video.
8. A method in a computing system for determining whether a query image is a logo, the method comprising:
receiving a query image that is to be evaluated to determine whether the image is a logo, wherein the image is comprised of a plurality of pixels, each pixel having an associated pixel value;
applying image tests comprising:
a first image test to determine whether an entropy of the pixel-values associated with each of the plurality of pixels of the query image exceeds a threshold entropy;
a second image test to determine whether an entropy of a plurality of edge crossings in the query image exceeds a threshold entropy; and
a third image test to determine whether a distribution of gradient magnitudes of the query image resembles a power function having a negative exponent;
associating each of the image tests with a weighting; and
identifying a likelihood that the query image is a logo based on the weighting of each of the image tests.
9. The method of claim 8 , further comprising, after receiving the query image, processing the query image by:
converting the query image to grayscale, and
resizing the query image.
10. A non-transitory computer-readable medium encoded with instructions that, when executed by a processor, perform a method to prevent logos from being searched within a catalog of images, the method comprising:
receiving a query image that is to be searched against a catalog of images, wherein the query image is comprised of a plurality of pixels;
evaluating pixel-values associated with each of the plurality of pixels of the query image to calculate a first factor indicative of whether the query image is a natural image;
evaluating a plurality of edge crossings in the query image to calculate a second factor indicative of whether the query image is a cartoon image;
evaluating gradient magnitudes of the query image to calculate a third factor indicative of whether the query image is a concept image, wherein evaluating the gradient magnitudes of the query image comprises evaluating whether a distribution of the gradient magnitudes resembles a power function having a negative exponent;
associating a weighting with each of the first, second, and third factors;
determining whether the query image is a logo based on the weighted first, second, and third factors;
using the query image to search within the catalog of images if the query image is not a logo; and
not using the query image to search within the catalog of images if the query image is a logo.
11. The non-transitory computer-readable medium of claim 10 , further encoded with instructions that when executed by the processor perform the method to prevent logos from being searched within a catalog of images, the method further comprising:
converting the query image to grayscale, and
resizing the query image.
12. The non-transitory computer-readable medium of claim 10 , wherein evaluating the pixel-values associated with each of the plurality of pixels of the query image comprises:
calculating the number of pixels in the query image having each of the pixel-values; and
calculating an entropy of the number of pixels in the query image having each of the pixel-values.
13. The non-transitory computer-readable medium of claim 10 , wherein evaluating the plurality of edge crossings in the query image comprises:
calculating the number of edge crossings in the query image; and
calculating an entropy of the number of edge crossings in the query image.
14. The non-transitory computer-readable medium of claim 10 , wherein evaluating the gradient magnitudes of the query image further comprises:
calculating the gradient magnitudes of the query image; and
generating the distribution of the gradient magnitudes in the query image.
15. The non-transitory computer-readable medium of claim 10 , wherein the evaluation of whether the distribution of gradient magnitudes resembles the power function having a negative exponent is based on calculating a linear approximation of the logarithm of the distribution and comparing the calculated linear approximation with the logarithm of the distribution.
16. The non-transitory computer-readable medium of claim 10 , wherein the query image is excerpted from a video.
17. A non-transitory computer-readable medium encoded with instructions that, when executed by a processor, perform a method for determining whether a query image is a logo, the method comprising:
receiving a query image that is to be evaluated to determine whether the image is a logo, wherein the image is comprised of a plurality of pixels, each pixel having an associated pixel value;
applying image tests comprising:
a first image test to determine whether an entropy of the pixel-values associated with each of the plurality of pixels of the query image exceeds a threshold entropy; and
a second image test to determine whether a distribution of gradient magnitudes of the query image resembles a power function having a negative exponent;
associating each of the image tests with a weighting; and
identifying a likelihood that the query image is a logo based on the weighting of each of the image tests.
18. The non-transitory computer-readable medium of claim 17 , further encoded with instructions that when executed by the processor perform the method for determining whether a query image is a logo, the method further comprising, after receiving the query image, processing the query image by:
converting the query image to grayscale, and
resizing the query image.