Optimizing a visual perspective of media

摘要

One or more signals are used to identify regions of interest of an image. The signals are applied to the image to generate one or more models that are based on the regions of interest. The models may present different perspectives of the image by emphasizing various features and focal points. The models may be ranked and displayed according to a scoring paradigm that is based on one or more signals. Multi-tiered feedback mechanisms allow for the collection of user intent and/or other forms of explicit input. Feedback associated to the models may be obtained and used to generate additional models that are based on one or more signals and the feedback. The feedback may also be stored and utilized for machine learning purposes.

主权项

1. A computer-implemented method comprising:
obtaining data defining an intended use of an image; determining a plurality of salient regions or a plurality of invariant regions of the image by applying a plurality of signals to the image; determining a confidence score for at least one salient region or at least one invariant region of the plurality of salient regions or the plurality of invariant regions, wherein the confidence score is based, at least in part on at least one of a size of an identifiable object in the image, a depth of a color or a luminance variation of the image, an existence of identifiable features of the identifiable object, and an image quality of the image; and generating a plurality of models, wherein individual models of the plurality of models focus on the at least one salient region or the at least one invariant region of the image, and wherein a selection of the at least one salient region or the at least one invariant region is based on, at least in part, the intended use and the confidence score for the at least one salient region or the at least one invariant region displaying a transformation of individual models of the plurality of models on an interface; receiving, at the interface, a gesture input indicating an intent directed toward one or more regions associated with at least one model; and in response to the gesture input, generating additional models based on the intent.