Back to results
US 2003/0194132 A1Published application
Picture region extraction method and device
AssigneeNEC Corporation
InventorTomoharu Kiyuna
Life of this patent
Estimating the term…
Abstract
The picture region extraction method coarse-grains this picture data space, calculates a coarse-grained empirical probability distribution, initializes parameters, calculates a coarse-grained conditional probability distribution and a class membership probability, updates the parameters, and calculates an evaluation function, each process being repeated until there is no change in the evaluation function, at which point, a picture region is extracted based on the class membership probability.
Claims20
A picture region extraction method, which extracts a desired region from picture data containing an image to be subjected to region extraction by classifying each pixel on the picture into a plurality of classes, comprising:
- a first step in which the data space constituted by all the attribute values that may be taken by said each pixel on said picture is divided into subspaces with a given resolution, a collection of pixels, each of which takes an attribute value in said each subspace, the average of the attribute values of said pixels, and the number of said pixels are retained to constitute a coarse-grained space;
- a second step in which the number of pixels in said each subspace is divided by the total number of pixels contained in said picture to calculate the coarse-grained empirical probability distribution in the coarse-grained data space;
- a third step in which the class parameter, the number of said classes, and the mixture ratio of said classes, which define the attributes of said each class, are initialized;
- a fourth step in which a conditional probability distribution under said class being specified is calculated from the class parameter that defines the attributes of said each class, and the conditional probability distribution under said class being specified is averaged within said each subspace to calculate a coarse-grained conditional probability distribution;
- a fifth step in which a class membership probability, which is the probability that each pixel constituting said picture belongs to said each class, is calculated by multiplying said class mixture ratio by said coarse-grained conditional probability distribution;
- a sixth step in which said class parameter and said class mixture ratio are updated so as to increase an evaluation function;
- a seventh step in which a coarse-grained log-likelihood is calculated as said evaluation function using said coarse-grained conditional probability distribution;
- an eighth step in which whether said evaluation function satisfies a given termination condition or not is examined;
- and
- a ninth step in which after said evaluation function satisfies said given termination condition, said class parameter and said class mixture ratio are retained, and the region each pixel belongs to is determined based on said class membership probability to extract said desired region,
- said fourth, fifth, sixth, seventh and eighth steps being repeated until said evaluation function satisfies said given condition.
8 more elements
A picture region extraction device, which extracts a desired region from picture data containing an image to be subjected to region extraction by classifying each pixel on the picture into a plurality of classes, comprising:
- an input device for reading picture data;
- a region coarse-graining device in which the data space constituted by all the attribute values that may be taken by said each pixel on said picture is divided into subspaces with a given resolution, a collection of pixels, each of which takes an attribute value in said each subspace, the average of the attribute values of said pixels, and the number of said pixels are retained to constitute a coarse-grained data space;
- a coarse-grained empirical probability distribution calculation device in which the number of pixels in said each subspace is divided by the total number of pixels contained in said picture to calculate the coarse-grained empirical distribution in the coarse-grained data space;
- a coarse-grained conditional probability distribution calculation device in which the class parameter, the number of said classes, and the mixture ratio of said classes, which define the attributes of said each class, are initialized, a conditional probability distribution under said class being specified is calculated from the class parameter that defines the attributes of said each class, and the conditional probability distribution under said class being specified is averaged within said each subspace to calculate the coarse-grained conditional probability distribution;
- a class membership probability calculation device in which a class membership probability, which is the probability that each pixel constituting said picture belongs to said each class, is calculated by said coarse-grained conditional probability distribution;
- a parameter update device in which said parameter is updated so that when a coarse-grained log-likelihood is used as said evaluation function, said evaluation function may be increased and when AIC, MDL, or Structural Risk is used as the evaluation function, the evaluation function may be decreased;
- an evaluation function calculation device in which said evaluation function is calculated using either said coarse-grained log-likelihood, said AIC, said MDL, or said Structural Risk;
- a region extraction device in which whether said evaluation function satisfies a given termination condition or not is examined, after said evaluation function satisfies said given termination condition, said parameter is retained, and the class said each pixel belongs to is determined based on said class membership probability to extract a region;
- and
- an output device for outputting the extracted region.
7 more elements
A program allowing a computer to execute functions for extracting a desired picture region from picture data containing an image to be subjected to region extraction by classifying each pixel on the picture into a plurality of classes, comprising:
- a first function in which the data space constituted by all the attribute values that may be taken by said each pixel constituting said picture is divided into subspaces with a given resolution, a collection of pixels, each of which takes an attribute value in said each subspace, the average of the attribute values of said pixels, and the number of said pixels are retained to constitute a coarse-grained space;
- a second function in which the number of pixels in said each subspace is divided by the total number of pixels contained in said picture to calculate the coarse-grained empirical probability distribution in the coarse-grained data space;
- a third function in which the class parameter, the number of said classes, and the mixture ratio of said classes, which define the attributes of said each class, are initialized;
- a fourth function in which a conditional probability distribution under said class being specified is calculated from the class parameter that defines the attributes of said each class, and the conditional probability distribution under said class being specified is averaged within said each subspace to calculate a coarse-grained conditional probability distribution;
- a fifth function in which a class membership probability, which is the probability that each pixel constituting said picture belongs to said each class, is calculated by multiplying said class mixture ratio by said coarse-grained conditional probability distribution;
- a sixth function in which said class parameter and said class mixture ratio are updated so as to increase an evaluation function;
- a seventh function in which a coarse-grained log-likelihood is calculated as said evaluation function using said coarse-grained conditional probability distribution;
- an eighth function in which whether said evaluation function satisfies a given termination condition or not is examined;
- and
- a ninth function in which after said evaluation function satisfies said given termination condition, said class parameter and said class mixture ratio are retained, and the region each pixel belongs to is determined based on said class membership probability to extract said desired region,
- said fourth, fifth, sixth, seventh, and eighth functions being repeated until said evaluation function satisfies said given condition.
8 more elements
Dependencies are inferred from the claim text.