Abstract: Text features corresponding to pieces of media content (e.g., images, audio, multimedia content, etc.) are extracted from media content sources. One or more text features (e.g., one or more words) for a piece of media content are extracted from text associated with the piece of media content and text feature vectors generated therefrom and used during subsequent searching. Additional low-level feature vectors may also be extracted from the piece of media content and used during the subsequent searching. Relevance feedback can also be received from a user(s) identifying the relevance of pieces of media content rendered to the user in response to his or her search request. The relevance feedback is logged and can be used in determining how to respond to subsequent search requests, such as by modifying feature vectors (e.g., text feature vectors) corresponding to the pieces of media content for which relevance feedback is received.

Abstract: In one aspect, the present disclosure describes a process for automatic artifact compensation in a digital representation of an image. The process includes detecting, by a processor, regions corresponding to facial images within the digital representation; locating, by the processor, red-eye regions within the detected regions; and automatically modifying, by the processor, the located red-eye regions to provide a modified image.

Abstract: Systems and methods for learning-based automatic commercial content detection are described. In one aspect, the systems and methods include a training component and an analyzing component. The training component trains a commercial content classification model using a kernel support vector machine. The analyzing component analyzes program data such as video and audio data using the commercial content classification model and one or more of single-side left neighborhood(s) and right neighborhood(s) of program data segments. Based on this analysis, each of the program data segments are classified as being commercial or non-commercial segments.

Abstract: Improved methods and apparatuses are provided for use in face detection. The methods and apparatuses significantly reduce the number of candidate windows within a digital image that need to be processed using more complex and/or time consuming face detection algorithms. The improved methods and apparatuses include a skin color filter and an adaptive non-face skipping scheme.

Abstract: Probabilistic retrospective event detection is described. In one aspect, event parameters are initialized to identify a number of events from a corpus of documents. Using a generative model, documents are determined to be associated with an event to detect representative events from the identified number of events.

Abstract: Systems and methods for learning-based automatic commercial content detection are described. In one aspect, program data is divided into multiple segments. The segments are analyzed to determine visual, audio, and context-based feature sets that differentiate commercial content from non-commercial content. The context-based features are a function of single-side left and/or right neighborhoods of segments of the multiple segments.

Abstract: Systems, engines, user interfaces, and methods allow a user to select a group of images, such as digital photographs, and assign to the group of images the name of a person who is represented in each of the images. The name is automatically propagated to the face of the person, each time the person's face occurs in an image. In one implementation, names and associations are shared between a browsing mode for viewing multiple images at once and a viewer mode, for viewing one image at a time. The browsing mode can provide a menu of candidate names for annotating a face in a single image of the viewer mode. Likewise, the viewer mode can provide annotated face information to the browser mode for facilitating name propagation. Identification of a person's face in multiple images can be accomplished not only by finding similarities in facial features but also by finding similarities in contextual features near the face in different images.

Abstract: Systems and methods to automatically edit a video to generate a video summary are described. In one aspect, sub-shots are extracted from the video. Importance measures are calculated for at least a portion of the extracted sub-shots. Respective relative distributions for sub-shots having relatively higher importance measures as compared to importance measures of other sub-shots are determined. Based on the determined relative distributions, sub-shots that do not exhibit a uniform distribution with respect to other sub-shots in the particular ones are dropped. The remaining sub-shots are connected with respective transitions to generate the video summary.

Abstract: A method and apparatus are provided for augmenting a language model with a class entity dictionary based on corrections made by a user. Under the method and apparatus, a user corrects an output that is based in part on the language model by replacing an output segment with a correct segment. The correct segment is added to a class of segments in the class entity dictionary and a probability of the correct segment given the class is estimated based on an n-gram probability associated with the output segment and an n-gram probability associated with the class. This estimated probability is then used to generate further outputs.

Abstract: A method and apparatus are provided for adapting a language model to a task-specific domain. Under the method and apparatus, the relative frequency of n-grams in a small training set (i.e. task-specific training data set) and the relative frequency of n-grams in a large training set (i.e. out-of-domain training data set) are used to weight a distribution count of n-grams in the large training set. The weighted distributions are then used to form a modified language model by identifying probabilities for n-grams from the weighted distributions.

Abstract: A method and apparatus are provided for adapting a language model to a task-specific domain. Under the method and apparatus, the relative frequency of n-grams in a small training set (i.e. task-specific training data set) and the relative frequency of n-grams in a large training set (i.e. out-of-domain training data set) are used to weight a distribution count of n-grams in the large training set. The weighted distributions are then used to form a modified language model by identifying probabilities for n-grams from the weighted distributions.

Abstract: The disclosed subject matter improves iterative results of content-based image retrieval (CBIR) using a bigram model to correlate relevance feedback. Specifically, multiple images are received responsive to multiple image search sessions. Relevance feedback is used to determine whether the received images are semantically relevant. A respective semantic correlation between each of at least one pair of the images is then estimated using respective bigram frequencies. The bigram frequencies are based on multiple search sessions in which each image of a pair of images is semantically relevant.

Abstract: The disclosed subject matter improves iterative results of content-based image retrieval (CBIR) using a bigram model to correlate relevance feedback. Specifically, multiple images are received responsive to multiple image search sessions. Relevance feedback is used to determine whether the received images are semantically relevant. A respective semantic correlation between each of at least one pair of the images is then estimated using respective bigram frequencies. The bigram frequencies are based on multiple search sessions in which each image of a pair of images is semantically relevant.

Abstract: Systems and methods for indexing and retrieving images are described herein. The systems and methods analyze an image to determine its texture moments. The pixels of the image are converted to gray scale. Textural attributes of the pixels are determined. The textural attributes are associated with the local texture of the pixels and are derived from coefficients of Discrete Fourier Transform associated with the pixels. Statistical values associated with the textural attributes of the pixels are calculated. The texture moments of the image are determined from the statistical value.

Abstract: A process for comparing two digital images is described. The process includes comparing texture moment data for the two images to provide a similarity index, combining the similarity index with other data to provide a similarity value and determining that the two images match when the similarity value exceeds a first threshold value.

Abstract: Systems and methods for shape registration are described. In one aspect, training shape vectors are generated from images in an image database. The training shape vectors identify landmark points associated with one or more object types. A distribution of shape in the training shape vectors is represented as a prior of tangent shape in tangent shape space. The prior of tangent shape is then incorporated into a unified Bayesian framework for shape registration.

Abstract: Face detection techniques are provided that use a multiple-stage face detection algorithm. An exemplary three-stage algorithm includes a first stage that applies linear-filtering to enhance detection performance by removing many non-face-like portions within an image, a second stage that uses a boosting chain that is adopted to combine boosting classifiers within a hierarchy “chain” structure, and a third stage that performs post-filtering using image pre-processing, SVM-filtering and color-filtering to refine the final face detection prediction. In certain further implementations, the face detection techniques include a two-level hierarchy in-plane pose estimator to provide a rapid multi-view face detector that further improves the accuracy and robustness of face detection.

Abstract: Improved methods and apparatuses are provided for use in face detection. The methods and apparatuses significantly reduce the number of candidate windows within a digital image that need to be processed using more complex and/or time consuming face detection algorithms. The improved methods and apparatuses include a skin color filter and an adaptive non-face skipping scheme.

Abstract: Systems and methods for annotating a face in a digital image are described. In one aspect, a probability model is trained by mapping one or more sets of sample facial features to corresponding names of individuals. A face from an input data set of at least one the digital image is then detected. Facial features are then automatically extracted from the detected face. A similarity measure is them modeled as a posterior probability that the facial features match a particular set of features identified in the probability model. The similarity measure is statistically learned. A name is then inferred as a function of the similarity measure. The face is then annotated with the name.

Abstract: Improvements are provided to effectively assess a user's face and head pose such that a computer or like device can track the user's attention towards a display device(s). Then the region of the display or graphical user interface that the user is turned towards can be automatically selected without requiring the user to provide further inputs. A frontal face detector is applied to detect the user's frontal face and then key facial points such as left/right eye center, left/right mouth corner, nose tip, etc., are detected by component detectors. The system then tracks the user's head by an image tracker and determines yaw, tilt and roll angle and other pose information of the user's head through a coarse to fine process according to key facial points and/or confidence outputs by pose estimator.