Abstract: A method for object classification by an electronic device is described. The method includes obtaining an image frame that includes an object. The method also includes determining samples from the image frame. Each of the samples represents a multidimensional feature vector. The method further includes adding the samples to a training set for the image frame. The method additionally includes pruning one or more samples from the training set to produce a pruned training set. One or more non-support vector negative samples are pruned first. One or more non-support vector positive samples are pruned second if necessary to avoid exceeding a sample number threshold. One or more support vector samples are pruned third if necessary to avoid exceeding the sample number threshold. The method also includes updating classifier model weights based on the pruned training set.

Abstract: A method performed by an electronic device is described. The method includes determining a local motion pattern by determining a set of local motion vectors within a region of interest between a previous frame and a current frame. The method also includes determining a global motion pattern by determining a set of global motion vectors between the previous frame and the current frame. The method further includes calculating a separation metric based on the local motion pattern and the global motion pattern. The separation metric indicates a motion difference between the local motion pattern and the global motion pattern. The method additionally includes tracking an object based on the separation metric.

Abstract: A method for tracking an object by an electronic device is described. The method includes detecting an object position in an initial frame to produce a detected object position. The method also includes measuring one or more landmark positions based on the detected object position or a predicted object position. The method further includes predicting the object position in a subsequent frame based on the one or more landmark positions. The method additionally includes determining whether object tracking is lost. The method also includes avoiding performing object detection for the subsequent frame in a case that object tracking is maintained.

Abstract: Embodiments include methods and systems for context-adaptive pixel processing based, in part, on a respective weighting-value for each pixel or a group of pixels. The weighting-values provide an indication as to which pixels are more pertinent to pixel processing computations. Computational resources and effort can be focused on pixels with higher weights, which are generally more pertinent for certain pixel processing determinations.

Abstract: A method for memory utilization by an electronic device is described. The method includes transferring a first portion of a first decision tree and a second portion of a second decision tree from a first memory to a cache memory. The first portion and second portion of each decision tree are stored contiguously in the first memory. The first decision tree and second decision tree are each associated with a different feature of an object detection algorithm. The method also includes reducing cache misses by traversing the first portion of the first decision tree and the second portion of the second decision tree in the cache memory based on an order of execution of the object detection algorithm.

Abstract: A method for object classification by an electronic device is described. The method includes obtaining an image frame that includes an object. The method also includes determining samples from the image frame. Each of the samples represents a multidimensional feature vector. The method further includes adding the samples to a training set for the image frame. The method additionally includes pruning one or more samples from the training set to produce a pruned training set. One or more non-support vector negative samples are pruned first. One or more non-support vector positive samples are pruned second if necessary to avoid exceeding a sample number threshold. One or more support vector samples are pruned third if necessary to avoid exceeding the sample number threshold. The method also includes updating classifier model weights based on the pruned training set.

Abstract: A method for obtaining structural information from a digital image by an electronic device is described. The method includes determining an iris position in a region of interest based on a gradient direction transform. Determining the iris position may include determining a first dimension position and a second dimension position corresponding to a maximum value in the transform space.

Abstract: A method performed by an electronic device is described. The method includes determining a haziness confidence level based on multiple modalities. The method also includes determining whether to perform an action based on the haziness confidence level. The method may include performing the action, including performing haziness reduction based on the haziness confidence level.

Abstract: A method includes capturing an image of a scene that includes a diagram. The method further includes applying functional block recognition rules to image data of the image to recognize functional blocks of the diagram. The functional blocks include at least a first functional block associated with a first computer operation. The method further includes determining whether the functional blocks comply with functional block syntax rules. A functional graph is computer-generated based on the functional blocks complying with the functional block syntax rules. The functional graph corresponds to the diagram, and the functional graph includes the functional blocks.

Abstract: Apparatus and methods for facial detection are disclosed. A plurality of images of an observed face is received for identification. Based at least on two or more selected images of the plurality of images, a template of the observed face is generated. In some embodiments, the template is a subspace generated based on feature vectors of the plurality of received images. A database of identities and corresponding facial data of known persons is searched based at least on the template of the observed face and the facial data of the known persons. One or more identities of the known persons are selected based at least on the search.

Abstract: A method for deformable expression detection is disclosed. For each pixel in a preprocessed image, a sign of a first directional gradient component and a sign of a second directional gradient component are combined to produce a combined sign. Each combined sign is coded into a coded value. An expression in an input image is detected based on the coded values.

Abstract: In general, this disclosure describes techniques for providing a gesture-based user interface. For example, according to some aspects of the disclosure, a user interface generally includes a camera and a computing device that identifies and tracks the motion of one or more fingertips of a user. In some examples, the user interface is configured to identify predefined gestures (e.g., patterns of motion) associated with certain motions of the user's fingertips. In another example, the user interface is configured to identify hand postures (e.g., patterns of showing up of fingertips). Accordingly, the user can interact with the computing device by performing the gestures.

Abstract: A method includes receiving first data defining a first bounding box for a first image of a sequence of images. The first bounding box corresponds to a region of interest including a tracked object. The method also includes receiving object tracking data for a second image of the sequence of images, the object tracking data defining a second bounding box. The second bounding box corresponds to the region of interest including the tracked object in the second image. The method further includes determining a similarity metric for first pixels within the first bounding box and search pixels within each of multiple search bounding boxes. Search coordinates of each of the search bounding boxes correspond to second coordinates of the second bounding box shifted in one or more directions. The method also includes determining a modified second bounding box based on the similarity metric.

Abstract: The techniques of this disclosure are directed to the feedback-based stereoscopic display of three-dimensional images, such as may be used for video telephony (VT) and human-machine interface (HMI) application. According to one example, a region of interest (ROI) of stereoscopically captured images may be automatically determined based on determining disparity for at least one pixel of the captured images are described herein. According to another example, a zero disparity plane (ZDP) for the presentation of a 3D representation of stereoscopically captured images may be determined based on an identified ROI. According to this example, the ROI may be automatically identified, or identified based on receipt of user input identifying the ROI.

Abstract: A method for memory utilization by an electronic device is described. The method includes transferring a first portion of a first decision tree and a second portion of a second decision tree from a first memory to a cache memory. The first portion and second portion of each decision tree are stored contiguously in the first memory. The first decision tree and second decision tree are each associated with a different feature of an object detection algorithm. The method also includes reducing cache misses by traversing the first portion of the first decision tree and the second portion of the second decision tree in the cache memory based on an order of execution of the object detection algorithm.

Abstract: A method for three-dimensional face generation is described. An inverse depth map is calculated based on a depth map and an inverted first matrix. The inverted first matrix is generated from two images in which pixels are aligned vertically and differ horizontally. The inverse depth map is normalized to correct for distortions in the depth map caused by image rectification. A three-dimensional face model is generated based on the inverse depth map and one of the two images.

Abstract: A method for deformable expression detection is disclosed. For each pixel in a preprocessed image, a sign of a first directional gradient component and a sign of a second directional gradient component are combined to produce a combined sign. Each combined sign is coded into a coded value. An expression in an input image is detected based on the coded values.

Abstract: A method includes capturing an image of a scene that includes a diagram. The method further includes applying functional block recognition rules to image data of the image to recognize functional blocks of the diagram. The functional blocks include at least a first functional block associated with a first computer operation. The method further includes determining whether the functional blocks comply with functional block syntax rules. A functional graph is computer-generated based on the functional blocks complying with the functional block syntax rules. The functional graph corresponds to the diagram, and the functional graph includes the functional blocks.

Abstract: Embodiments include methods and systems for context-adaptive pixel processing based, in part, on a respective weighting-value for each pixel or a group of pixels. The weighting-values provide an indication as to which pixels are more pertinent to pixel processing computations. Computational resources and effort can be focused on pixels with higher weights, which are generally more pertinent for certain pixel processing determinations.

Abstract: A method for face detection is disclosed. The method includes evaluating a scanning window using a first weak classifier in a first stage classifier. The method also includes evaluating the scanning window using a second weak classifier in the first stage classifier based on the evaluation using the first weak classifier.