Microsoft Technology Licensing, LLC

Other

Microsoft Corporation

Chuang Gu

Ming-Chieh Lee

Stacey Spears

Yaming He

Chun-Wei Chan

William Chen

Florin Folta

Thomas W Holcomb

Wei-ge Chen

Chad Gibson

King Wei Hor

David Alles

Anil Murching

Chih-Lung Lin

Sanjeev Mehrotra

Minesh G Shilotri

Various embodiments related to the generation and provision of media metadata are disclosed. For example, one disclosed embodiment provides a computing device having a logic subsystem configured to execute instructions, and a data holding subsystem comprising instructions stored thereon that are executable by the processor to receive an input of a video and/or audio content item, and to compare the content item to one or more object descriptors each representing an object for locating within the content item to locate instances of one or more of the objects in the content item. The instructions are further executable to generate metadata for each object located in the video content item, and to receive a validating user input related to whether the metadata generated for a selected object is correct.

Generation and provision of media metadata

A multiple bitrate (MBR) video encoding management tool utilizes available processing units for parallel MBR video encoding. Instead of focusing only on multi-threading of encoding tasks for a single picture or group of pictures (GOP), the management tool parallelizes the encoding of multiple GOPs between different processing units and/or different computing systems. With this parallel MBR video encoding architecture, different GOPs can be encoded in parallel. To facilitate such parallel encoding, data dependencies between GOPs are removed. The management tool can adjust the number of GOPs to encode in parallel on a computing system so as to favor parallelism of encoding for different GOPs at the expense of parallelism of encoding inside a GOP, or vice versa, and thereby set a suitable balance between encoding latency and throughput.

Parallel multiple bitrate video encoding to reduce latency and dependences between groups of pictures

Various examples are disclosed herein that relate to staged element classification. For example, one disclosed example provides a method of classifying elements by forming elements for classification into a plurality of first-level sets in a first stage, generating primary groups within the first-level sets based on element similarity, forming a plurality of second-level sets from the first-level sets in a second stage, generating secondary groups within the second-level sets based on element similarity, and merging a plurality of the primary and/or secondary groups based on element similarity.

Staged element classification

The subject disclosure relates to face recognition in video. Face detection data in frames of input data are used to generate face galleries, which are labeled and used in recognizing faces throughout the video. Metadata that associates the video frame and the face are generated and maintained for subsequent identification. Faces other than those found by face detection may be found by face tracking, in which facial landmarks found by the face detection are used to track a face over previous and/or subsequent video frames. Once generated, the maintained metadata may be accessed to efficiently determine the identity of a person corresponding to a viewer-selected face.

Face recognition in video content

A video encoding system encodes video streams for multiple bit rate video streaming using an approach that permits the encoded bit rate to vary subject to a peak bit rate and average bit rate constraints for higher quality streams, while a bottom bit rate stream is encoded to achieve a constant chunk rate. The video encoding system also dynamically decides an encoding resolution for segments of the multiple bit rate video streams that varies with video complexity so as to achieve a better visual experience for multiple bit rate streaming.

Multiple bit rate video encoding using variable bit rate and dynamic resolution for adaptive video streaming

A video encoder uses previously calculated motion information for inter frame coding to achieve faster computation speed for video compression. In a multi bit rate application, motion information produced by motion estimation for inter frame coding of a compressed video bit stream at one bit rate is passed on to a subsequent encoding of the video at a lower bit rate. The video encoder chooses to use the previously calculated motion information for inter frame coding at the lower bit rate if the video resolution is unchanged. A multi core motion information pre-calculation produces motion information prior to encoding by dividing motion estimation of each inter frame to separate CPU cores.

Video encoding using previously calculated motion information

A video encoding system encodes video streams for multiple bit rate video streaming using an approach that permits the encoded resolution to vary based, at least in part, on motion complexity. The video encoding system dynamically decides an encoding resolution for segments of the multiple bit rate video streams that varies with video complexity so as to achieve a better visual experience for multiple bit rate streaming. Motion complexity may be considered separately, or along with spatial complexity, in making the resolution decision.

Motion based dynamic resolution multiple bit rate video encoding

Image processing in mobile devices is optimized by combining at least two of the color conversion, rotation, and scaling operations. Received images, such as still images or frames of video stream, are subjected to a combined transformation after decoding, where each pixel is color converted (e.g. from YUV to RGB), rotated, and scaled as needed. By combining two or three of the processes into one, read/write operations consuming significant processing and memory resources are reduced enabling processing of higher resolution images and/or power and processing resource savings.

Rotation and scaling optimization for mobile devices

A semantic object tracking method tracks general semantic objects with multiple non-rigid motion, disconnected components and multiple colors throughout a vector image sequence. The method accurately tracks these general semantic objects by spatially segmenting image regions from a current frame and then classifying these regions as to which semantic object they originated from in the previous frame. To classify each region, the method performs a region based motion estimation between each spatially segmented region and the previous frame to compute the position of a predicted region in the previous frame. The method then classifies each region in the current frame as being part of a semantic object based on which semantic object in the previous frame contains the most overlapping points of the predicted region. Using this method, each region in the current image is tracked to one semantic object from the previous frame, with no gaps or overlaps. The method propagates few or no errors because it projects regions into a frame where the semantic object boundaries are previously computed rather than trying to project and adjust a boundary in a frame where the object's boundary is unknown.

Tracking semantic objects in vector image sequences

A semantic object tracking method tracks general semantic objects with multiple non-rigid motion, disconnected components and multiple colors throughout a vector image sequence. The method accurately tracks these general semantic objects by spatially segmenting image regions from a current frame and then classifying these regions as to which semantic object they originated from in the previous frame. To classify each region, the method perform a region based motion estimation between each spatially segmented region and the previous frame to computed the position of a predicted region in the previous frame. The method then classifies each region in the current frame as being part of a semantic object based on which semantic object in the previous frame contains the most overlapping points of the predicted region. Using this method, each region in the current image is tracked to one semantic object from the previous frame, with no gaps or overlaps. The method propagates few or no errors because it projects regions into a frame where the semantic object boundaries are previously computed rather than trying to project and adjust a boundary in a frame where the object's boundary is unknown.

Region extraction in vector images

A semantic video object extraction system using mathematical morphology and perspective motion modeling. A user indicates a rough outline around an image feature of interest for a first frame in a video sequence. Without further user assistance, the rough outline is processed by a morphological segmentation tool to snap the rough outline into a precise boundary surrounding the image feature. Motion modeling is performed on the image feature to track its movement into a subsequent video frame. The motion model is applied to the precise boundary to warp the precise outline into a new rough outline for the image feature in the subsequent video frame. This new rough outline is then snapped to locate a new precise boundary. Automatic processing is repeated for subsequent video frames.

Semantic video object segmentation and tracking

A human speech detection method detects pure-speech signals in an audio signal containing a mixture of pure-speech and non-speech or mixed-speech signals. The method accurately detects the pure-speech signals by computing a novel Valley Percentage feature from the audio signal and then classifying the audio signals into pure-speech and non-speech (or mixed-speech) classifications. The Valley Percentage is a measurement of the low energy parts of the audio signal (the valley) in comparison to the high energy parts of the audio signal (the mountain). To classify the audio signal, the method performs a threshold decision on the value of the Valley Percentage. Using a binary mask, a high Valley Percentage is classified as pure-speech and a low Valley Percentage is classified as non-speech (or mixed-speech). The method further employs morphological filters to improve the accuracy of human speech detection. Before detection, a morphological closing filter may be employed to eliminate unwanted noise from the audio signal. After detection, a combination of morphological closing and opening filters may be employed to remove aberrant pure-speech and non-speech classifications from the binary mask resulting from impulsive audio signals in order to more accurately detect the boundaries between the pure-speech and non-speech portions of the audio signal. A number of parameters may be employed by the method to further improve the accuracy of human speech detection. For implementation in supervised digital audio signal applications, these parameters may be optimized by training the application a priori. For implementation in an unsupervised environment, adaptive determination of these parameters is also possible.

Morphological pure speech detection using valley percentage

Homogeneous moving objects of arbitrary shapes are segmented and tracked with respect to the motion of the objects. In an intraframe mode of operation, a segmentation method includes obtaining a motion representation of corresponding pixels in the selected video image frame and a preceding video image frame to form motion-segmented video image features. Video image features are also segmented according to their spatial image characteristics (e.g., color) to form spatially-segmented video image features. Finally, the video image features are jointly segmented as a weighted combination of the motion-segmented video image features and the spatially-segmented video image features. The joint motion and spatial segmentation of image features provides enhanced accuracy in representing moving image features. This enhanced accuracy is particularly beneficial because the motion of image features is a significant display characteristic for human observers.

Growing community of inventors

Bellevue, WA, United States of America

Chuang Gu