IMAGE CLASSIFICATION
First Claim
1. A method for classifying photo and graphic images using features generated from image statistics, the method comprising:
- determining one or more classifying features of an image, comprising at least one of;
using a gradient magnitude-orientation histogram to determine one or more classifying features of the image, comprising;
generating a gradient magnitude-orientation histogram for the image; and
measuring a difference between an energy value of a lower frequency color band and an energy value of a higher frequency color band using the gradient magnitude-orientation histogram;
using multidimensional independent histograms for respective red, green, and blue color bands to determine one or more classifying features of the image, comprising;
generating respective multidimensional independent histograms corresponding to the red, green, and blue color bands for the image;
using the multidimensional independent histograms for the image to determine one or more of;
a statistical mean value;
a statistical variance value;
a statistical skewness value; and
a statistical kurtosis value; and
using the classifying features in a boosting decision tree for image classification, comprising at least one of;
training a boosting decision tree for image classification using the classifying features; and
classifying the image using the classifying features in a trained boosting decision tree.
2 Assignments
0 Petitions
Accused Products
Abstract
Images are classified as photos (e.g., natural photographs) or graphics (e.g., cartoons, synthetically generated images), such that when searched (online) with a filter, an image database returns images corresponding to the filter criteria (e.g., either photos or graphics will be returned). A set of image statistics pertaining to various visual cues (e.g., color, texture, shape) are identified in classifying the images. These image statistics, combined with pre-tagged image metadata defining an image as either a graphic or a photo, may be used to train a boosting decision tree. The trained boosting decision tree may be used to classify additional images as graphics or photos based on image statistics determined for the additional images.
64 Citations
20 Claims
-
1. A method for classifying photo and graphic images using features generated from image statistics, the method comprising:
-
determining one or more classifying features of an image, comprising at least one of; using a gradient magnitude-orientation histogram to determine one or more classifying features of the image, comprising; generating a gradient magnitude-orientation histogram for the image; and measuring a difference between an energy value of a lower frequency color band and an energy value of a higher frequency color band using the gradient magnitude-orientation histogram; using multidimensional independent histograms for respective red, green, and blue color bands to determine one or more classifying features of the image, comprising; generating respective multidimensional independent histograms corresponding to the red, green, and blue color bands for the image; using the multidimensional independent histograms for the image to determine one or more of; a statistical mean value; a statistical variance value; a statistical skewness value; and a statistical kurtosis value; and using the classifying features in a boosting decision tree for image classification, comprising at least one of; training a boosting decision tree for image classification using the classifying features; and classifying the image using the classifying features in a trained boosting decision tree. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
-
-
17. A system configured to classify photo and graphic images using features generated from image statistics, the system comprising:
-
An image classifying feature determiner comprising; a gradient magnitude orientation histogram generator configured to generate a gradient magnitude orientation histogram of an image; a gradient magnitude orientation histogram energy value determiner, configured to measure a difference between an energy value of a lower frequency color band and an energy value of a higher frequency color band of the gradient magnitude orientation histogram; a color histogram generator configured to generate independent histograms for respective red, green, and blue color bands of the image; and a color histogram statistical values determiner configured to determine statistical values of a color histogram, the values comprising at least one of; a statistical discreteness value; a statistical mean value; a statistical variance value; a statistical skewness value; and a statistical kurtosis value; and a boosting decision tree trainer configured to train an image classification boosting decision tree for graphic and photo image classification. - View Dependent Claims (18, 19)
-
-
20. A method for classifying photo and graphic images using features generated from image statistics, the method comprising:
-
determining one or more classifying features of an image, comprising at least one of; using a gradient magnitude-orientation histogram to determine one or more classifying features of the image, comprising; generating a gradient magnitude-orientation histogram for the image; measuring a difference between an energy value of a lower frequency color band and an energy value of a higher frequency color band using the gradient magnitude-orientation histogram; and determining an entropy value of the gradient magnitude-orientation histogram; using a joint multidimensional red-green-blue histogram of the image to determine one or more classifying features of the image comprising; generating a joint multidimensional red-green-blue histogram of the image; and determining an entropy value for the red-green-blue histogram of the image; using multidimensional independent histograms for respective red, green, and blue color bands to determine one or more classifying features of the image, comprising; generating respective multidimensional independent histograms corresponding to the red, green, and blue color bands for the image; using the multidimensional independent histograms for the image to determine one or more of; a statistical discreteness value a statistical mean value; a statistical variance value; a statistical skewness value; and a statistical kurtosis value; using a multidimensional local binary pattern texture histogram of the image to determine one or more classifying features of the image comprising; generating a multidimensional local binary pattern texture histogram of the image; and determining an entropy value for the texture histogram of the image; using a canny edge detector on the image to determine one or more classifying features of the image, comprising; determining a total number of edges in the image; and determining an average length of edges in the image; using a spatial correlogram of gray level pixels of the image to determine one or more classifying features of the image, comprising; determining an average skewness of respective slices of the spatial correlogram; and determining an average variance ratio, the variance ratio comprising a ratio of variance of respective slices to radius of a symmetric range for the spatial correlogram; and using the classifying features in a boosting decision tree for image classification, comprising at least one of; training a boosting decision tree for image classification using the classifying features; and classifying the image using the classifying features in a trained boosting decision tree.
-
Specification