Intelligent Video Thumbnail Selection and Generation

摘要

In accordance with one embodiment, an intelligent video thumbnail selection and generation tool may select a relevant and visually stimulating image from a video file and generate a thumbnail including the image. The image may be selected by computing a relevancy metric for an image in the file based on one or more selected relevant features, and comparing that relevancy metric with the metric of at least one other image in the file. In another embodiment, a series of images in a video file may be divided into shots. One of the shots may be selected based on a shot relevancy metric and a key image from the shot may be selected as a thumbnail based on a key image relevancy metric, where the shot relevancy metric and the key image relevancy metrics may be computed based on one or more relevant content features.

主权项

1. A method for recommending a thumbnail image representative of a collection of images, the method comprising:
obtaining a relevancy metric for each image of the collection of images, wherein the relevancy metric is computed based on at least one image characteristic selected from the group consisting of:
a size of a face in the image, wherein the relevancy metric is influenced in a first direction more when the face is large than when the face is small;a number of eyes in the image, wherein the relevancy metric is influenced in the first direction more when the number of eyes is large than when the number of eyes is small;a number of open eyes in the image, wherein the relevancy metric is influenced in the first direction more when the number of open eyes is large than when the number of open eyes is small;an object or face in the image, wherein the relevancy metric is influenced in the first direction more when the object or face is identified as corresponding to a database entry than when the object or face is unidentified; anda number of skin-colored pixels in the image, wherein the relevancy metric is influenced in the first direction more when the number of skin-colored pixels is large than when the number of skin-colored pixels is small; identifying a subset of the collection of images that each have a relevancy metric influenced in the first direction by an amount that satisfies a predetermined threshold; and transmitting for presentation through a user interface the identified subset of the collection of images.