Abstract: A computer-implemented method includes receiving a video that includes multiple frames. The method further includes identifying a start time and an end time of each action in the video based on application of one or more of an audio classifier, an RGB classifier, and a motion classifier. The method further includes identifying video segments from the video that include frames between the start time and the end time for each action in the video. The method further includes generating a confidence score for each of the video segments based on a probability that a corresponding action corresponds to one or more of a set of predetermined actions. The method further includes selecting a subset of the video segments based on the confidence score for each of the video segments.

Abstract: A computer-implemented method includes receiving a video that includes multiple frames. The method further includes identifying a start time and an end time of each action in the video based on application of one or more of an audio classifier, an RGB classifier, and a motion classifier. The method further includes identifying video segments from the video that include frames between the start time and the end time for each action in the video. The method further includes generating a confidence score for each of the video segments based on a probability that a corresponding action corresponds to one or more of a set of predetermined actions. The method further includes selecting a subset of the video segments based on the confidence score for each of the video segments.

Abstract: A computer-implemented method includes receiving a video that includes multiple frames. The method further includes identifying a start time and an end time of each action in the video based on application of one or more of an audio classifier, an RGB classifier, and a motion classifier. The method further includes identifying video segments from the video that include frames between the start time and the end time for each action in the video. The method further includes generating a confidence score for each of the video segments based on a probability that a corresponding action corresponds to one or more of a set of predetermined actions. The method further includes selecting a subset of the video segments based on the confidence score for each of the video segments.

Abstract: A computer-implemented method includes receiving a video that includes multiple frames. The method further includes identifying a start time and an end time of each action in the video based on application of one or more of an audio classifier, an RGB classifier, and a motion classifier. The method further includes identifying video segments from the video that include frames between the start time and the end time for each action in the video. The method further includes generating a confidence score for each of the video segments based on a probability that a corresponding action corresponds to one or more of a set of predetermined actions. The method further includes selecting a subset of the video segments based on the confidence score for each of the video segments.

Abstract: A computer-implemented method includes receiving a video that includes multiple frames. The method further includes identifying a start time and an end time of each action in the video based on application of one or more of an audio classifier, an RGB classifier, and a motion classifier. The method further includes identifying video segments from the video that include frames between the start time and the end time for each action in the video. The method further includes generating a confidence score for each of the video segments based on a probability that a corresponding action corresponds to one or more of a set of predetermined actions. The method further includes selecting a subset of the video segments based on the confidence score for each of the video segments.

Abstract: A computer-implemented method includes receiving a video that includes multiple frames. The method further includes identifying a start time and an end time of each action in the video based on application of one or more of an audio classifier, an RGB classifier, and a motion classifier. The method further includes identifying video segments from the video that include frames between the start time and the end time for each action in the video. The method further includes generating a confidence score for each of the video segments based on a probability that a corresponding action corresponds to one or more of a set of predetermined actions. The method further includes selecting a subset of the video segments based on the confidence score for each of the video segments.

Abstract: An improved ghost-artifact detection and removal method for high-dynamic range (HDR) image creation and related apparatus are described. A binary ghost map is first generated for each image of the multiple input images by a ghost-artifact detection process, where one of the binary values indicate ghost pixels and the other indicates non-ghost pixels. Each binary ghost map is smoothed to generate a continuous-tone ghost map, by changing the pixel value of each non-ghost pixel of the binary map to a ghost value between the two binary values. The ghost value is calculated using a monotonous function of the distance between the non-ghost pixel of the binary map and the nearest ghost pixel. Ghost pixels in the binary ghost map are kept as fully ghost pixel in the continuous-tone ghost map. This method helps to reduce visibility of artifacts at ghost boundaries without losing small detected ghost regions.

Abstract: An improved method for generating high dynamic range images by modifying the weight function used in conventional methods to weigh pixel values when combining multiple images in an input image set. Different weight functions, as functions of intensity, are used for different images (brackets). For darker brackets, a region near the high end of the intensity range is given higher weight values than a symmetrical region near the lower end of the intensity range; for brighter brackets, a region near the lower end of the intensity range is given higher weight values than a symmetrical region near the higher end of the intensity range. The weight function for the middle brackets are kept symmetrical but its function is non-zero at the two end points of the intensity range. This is effective for reducing chromatic artifacts for under- or over-exposed areas, in particular when ghost artifact removal is incorporated.

Abstract: In high dynamic range (HDR) image creation, a ghost artifact detection method divides the images (brackets) into multiple tiles, and selects one bracket for each tile as the local reference bracket. The local reference brackets are selected via optimization of an objective function which includes both a component that measures exposure quality of individual tiles and a component that measures correlation between neighboring tiles. The minimization can be realized by constructing a graph for the objective function and calculating a minimum cut of the graph using a graph cut algorithm. Graph examples for three and four image sets are given. Ghost artifact detection is then performed on a tile-by-tile basis by using the local reference bracket for each tile. Ghost maps are generated this way and used for HDR image creation. This method minimizes artifacts due to inconsistencies in local reference bracket selection in areas involved in ghost-inducing objects.

Abstract: An improved method for generating high dynamic range images by modifying the weight function used in conventional methods to weigh pixel values when combining multiple images in an input image set. Different weight functions, as functions of intensity, are used for different images (brackets). For darker brackets, a region near the high end of the intensity range is given higher weight values than a symmetrical region near the lower end of the intensity range; for brighter brackets, a region near the lower end of the intensity range is given higher weight values than a symmetrical region near the higher end of the intensity range. The weight function for the middle brackets are kept symmetrical but its function is non-zero at the two end points of the intensity range. This is effective for reducing chromatic artifacts for under- or over-exposed areas, in particular when ghost artifact removal is incorporated.

Abstract: An improved ghost-artifact detection and removal method for high-dynamic range (HDR) image creation and related apparatus are described. A binary ghost map is first generated for each image of the multiple input images by a ghost-artifact detection process, where one of the binary values indicate ghost pixels and the other indicates non-ghost pixels. Each binary ghost map is smoothed to generate a continuous-tone ghost map, by changing the pixel value of each non-ghost pixel of the binary map to a ghost value between the two binary values. The ghost value is calculated using a monotonous function of the distance between the non-ghost pixel of the binary map and the nearest ghost pixel. Ghost pixels in the binary ghost map are kept as fully ghost pixel in the continuous-tone ghost map. This method helps to reduce visibility of artifacts at ghost boundaries without losing small detected ghost regions.

Abstract: In high dynamic range (HDR) image creation, a ghost artifact detection method divides the images (brackets) into multiple tiles, and selects one bracket for each tile as the local reference bracket. The local reference brackets are selected via optimization of an objective function which includes both a component that measures exposure quality of individual tiles and a component that measures correlation between neighboring tiles. The minimization can be realized by constructing a graph for the objective function and calculating a minimum cut of the graph using a graph cut algorithm. Graph examples for three and four image sets are given. Ghost artifact detection is then performed on a tile-by-tile basis by using the local reference bracket for each tile. Ghost maps are generated this way and used for HDR image creation. This method minimizes artifacts due to inconsistencies in local reference bracket selection in areas involved in ghost-inducing objects.