System and method of analyzing images using a hierarchical set of models

Number: 10163036
Published: 2018-12-25
Filed: 2016-09-22
Assignee: Disney Enterprises, Inc.
Inventors: Lehrmann; Andreas et al.
CPC: G06F18/28; G06V10/7625; G06V30/196; G06F18/231; G06V10/772; G06F18/21; G06F18/24; G06V10/768
Verdict: Set aside generic image classifier/no creative hook
Source: Google Patents · FreePatentsOnline

Abstract

One or more image parameters of an image may be analyzed using a hierarchical set of models. Executing individual models in the set of models may generate outputs from analysis of different image parameters of the image. Inputs of one or more of the models may be conditioned on a set of outputs derived from one or more preceding model in the hierarchy.

Background

FIELD OF THE DISCLOSURE(1) This disclosure relates to analyzing images using a hierarchical set of models.BACKGROUND(2) Understanding a visual scene portrayed in an image may employ techniques including one or more of detecting scene features, recognizing objects from the detected features (e.g., identifying, categorizing, and/or other techniques for object recognition), determining locations of objects within the scene, and/or determining other information associated with the scene. Contextual models for understanding a scene may attempt to build various models of various forms for recognition. A simplest among those may look at label co-occurrence or exclusion among object categories in a given image. Others may look at enhancement or inhibition of detections using both co-occurrence and spatial local contextual relations, for example, through the use of structured image labeling, visual phrases, or discovered object groups, and/or to order detectors such that weaker detectors may benefit from stronger ones. Some models may look at context across granularities, for example, using texture patches to enhance performance of object detectors.SUMMARY(3) One aspect of the disclosure relates to a system configured for analyzing images using a hierarchical set of models. One or more of the models may be conditioned on a set of outputs derived from one or more preceding models such that output from an individual model may require execution of a limited set of the models. By way of n

Claims

1. A system configured for analyzing images using a hierarchical set of models, the system comprising: one or more physical processors configured by machine-readable instructions to: analyze an image by executing individual models in the hierarchical set of models, wherein executing the individual models in the set of models generates outputs from analysis of different image parameters of the image, wherein inputs of one or more of the models are solely conditioned on a complete set of outputs derived from a preceding model, the set of models includes: a first model configured to generate outputs from analysis of a first image parameter; a second model configured to generate outputs from analysis of a second image parameter, the execution of the second model being conditioned on outputs of the first model; a third model configured to generate outputs from analysis of a third image parameter, execution of the third model being conditioned on outputs of the second model; wherein the generation of outputs by the first model is independent from any outputs of the second model and third model; and wherein the generation of outputs by the second model is conditioned on outputs of the first model and is independent from outputs of the third model. 12. A method of analyzing images using a hierarchical set of models, the method being implemented in a computer system comprising one or more physical processors and storage media storing machine-readable instructions, the method comprising: analyzing an image by executing individual models in the hierarchical set of models, wherein executing the individual models in the set of models generates outputs from analysis of different image parameters of the image, wherein inputs of one or more of the models are solely conditioned on a complete set of outputs derived from a preceding model, the set of models includes: a first model configured to generate outputs from analysis of a first image parameter; a second model configured to generate outputs from analysis of a second image parameter, the execution of the second model being conditioned on outputs of the first model; and a third model configured to generate outputs from analysis of a third image parameter, execution of the third model being conditioned on outputs of the second model; wherein the generation of outputs by the first model is independent from any outputs of the second model and third model; and wherein the generation of outputs by the second model is conditioned on outputs of the first model and is independent from outputs of the third model.