Abstract: Techniques are disclosed for image feature representation. The techniques exhibit discriminative power that can be used in any number of classification tasks, and are particularly effective with respect to fine-grained image classification tasks. In an embodiment, a given image to be classified is divided into image patches. A vector is generated for each image patch. Each image patch vector is compared to the Gaussian mixture components (each mixture component is also a vector) of a Gaussian Mixture Model (GMM). Each such comparison generates a similarity score for each image patch vector. For each Gaussian mixture component, the image patch vectors associated with a similarity score that is too low are eliminated. The selectively pooled vectors from all the Gaussian mixture components are then concatenated to form the final image feature vector, which can be provided to a classifier so the given input image can be properly categorized.

Abstract: Certain embodiments involve using labels to track high-frequency offsets for patch-matching. For example, a processor identifies an offset between a first source image patch and a first target image patch. If the first source image patch and the first target image patch are sufficiently similar, the processor updates a data structure to include a label specifying the offset. The processor associates, via the data structure, the first source image patch with the label. The processor subsequently selects certain high-frequency offsets, including the identified offset, from frequently occurring offsets in the data structure. The processor uses these offsets to identify a second target image patch, which is located at the identified offset from a second source image patch. The processor associates, via the data structure, the second source image patch with the identified offset based on a sufficient similarity between the second source image patch and the second target image patch.

Abstract: Embodiments of the present invention relate to automatically identifying structures of a music stream. A segment structure may be generated that visually indicates repeating segments of a music stream. To generate a segment structure, a feature that corresponds to a music attribute from a waveform corresponding to the music stream is extracted from a waveform, such as an input signal. Utilizing a signal segmentation algorithm, such as a Variable Markov Oracle (VMO) algorithm, a symbolized signal, such as a VMO structure, is generated. From the symbolized signal, a matrix is generated. The matrix may be, for instance, a VMO-SSM. A segment structure is then generated from the matrix. The segment structure illustrates a segmentation of the music stream and the segments that are repetitive.

Abstract: Techniques are disclosed for the simulation of Internet of Things (IoT) device messages to control the generation and rendering of digital signage on a display module. A methodology implementing the techniques according to an embodiment includes intercepting messages that are transmitted from the devices to the display modules, and extracting or otherwise accessing sensor data from the intercepted messages. The sensor data can be provided to a user interface configured to allow a user to edit or manipulate the sensor data. A modified or transformed message is generated, based on the intercepted message and the edited sensor data. The modified or transformed message is transmitted to the display module, which is configured to render digital signage based on the content of the message. These techniques enable the user to visualize and evaluate alternative signage renderings in a relatively convenient manner by manipulating the sensor data reported from the IoT devices.

Abstract: Photometric stabilization for time-compressed video is described. Initially, video content captured by a video capturing device is time-compressed by selecting a subset of frames from the video content according to a frame sampling technique. Photometric characteristics are then stabilized across the frames of the time-compressed video. This involves determining correspondences of pixels in adjacent frames of the time-compressed video. Photometric transformations are then determined that describe how photometric characteristics (e.g., one or both of luminance and chrominance) change between the adjacent frames, given movement of objects through the captured scene. Based on the determined photometric transformations, filters are computed for smoothing photometric characteristic changes across the time-compressed video. Photometrically stabilized time-compressed video is generated from the time-compressed video by using the filters to smooth the photometric characteristic changes.

Abstract: Techniques for converting mechanical markings on hardcopy textual content into digital annotations in a digital document file. In accordance with some embodiments, the techniques include identifying at least one block of text in a digital (scanned) image of a hardcopy document, and identifying at least one mechanical marking in the digital image of the hardcopy document. The mechanical marking, such as an underline, strike-through, highlight or circle, covers or lies adjacent to the corresponding block of text. An annotated digital document file is generated from the digital image of the hardcopy document. The annotated digital document file includes computer-executable instructions representing the original text of the hardcopy document and at least one annotation corresponding to the mechanical marking in the hardcopy document.

Abstract: Techniques are provided for synchronizing comments between a platform-dependent source document and a platform-independent shared document created from the source document. The shared document can accept feedback, edits, annotations, and other comments (all generally referred to herein as comments) from a variety of platforms and collaborators. The comments entered in the shared document automatically become part of, or otherwise become visible from, the source document. In an embodiment, the techniques are implemented as a computer-driven methodology that includes creating a logical link between the source and shared documents. This link is based on unique IDs of the respective documents, and is used to determine that reviewer comments have been made in the shared document. In response to the source document being open on a given source tool, periodic polling for reviewer comments is carried out and any new comments made in the shared document are displayed in the source tool.

Abstract: Facilitating analysis of user interface gesture patterns is described. In example implementations, a computing device acquires data that describes a user interface gesture pattern, such as finger movements on a touchscreen, in terms of pixels of a visual display. A repetitive arrangement of polygons, such as a grid of rectangles, is logically overlaid on the visual display. The computing device transforms the pixel-based data into polygon-based data that represents the gesture pattern in terms of polygons traversed by the gesture pattern. The computing device also converts the polygon-based data into text-based data such that the gesture pattern is represented by textual characters, such as a text string. The text string can include, for instance, a list of polygons traversed by the gesture pattern. The text-based data is forwarded to a service that can efficiently analyze relationships that may exist among multiple gesture patterns across multiple devices or end users.

Abstract: Embodiments of an event-driven resource management technique may enable the management of cluster resources at a sub-computer level (e.g., at the thread level) and the decomposition of jobs at an atomic (task) level. A job queue may request a resource for a job from a resource manager, which may locate a resource in a resource list and grant the resource to the job queue. After the resource is granted, the job queue sends the job to the resource, on which the job may be partitioned into tasks and from which additional resources may be requested from the resource manager. The resource manager may locate additional resources in the list and grant the resources to the resource. The resource sends the tasks to the granted resources for execution. As resources complete their tasks, the resource manager is informed so that the status of the resources in the list can be updated.

Abstract: A system, to persist state in a dynamic electronic form, may include a communications module to receive a request to certify a template associated with an electronic form, an attribute detector to determine that the template is associated with a restore state attribute and a certification module. The certification module may be configured to selectively grant the request to certify the template based on the restore state attribute.

Abstract: Systems and methods are disclosed for generating a virtual reality environment from an electronic document by mapping image and text content extracted from the electronic document to a virtual reality module. A computing system performs a process to analyze an electronic document to identify the document elements that occur within the electronic document. The document elements are parsed to identify candidate images that have minimum image characteristics suitable for virtual reality. One or more of the candidate images are inserted into the virtual reality module. Text elements related to the candidate images are also inserted into the virtual reality module. Upon execution of the virtual reality module by a virtual reality viewing device, the text and image elements from the electronic document are presented in a virtual reality environment to the viewer for non-linear exploration and reading.

Abstract: In various implementations, methods and systems are disclosed for accurately selecting a targeted portion of a digital image. In one embodiment, a selection cursor having a central and a peripheral region is provided. The central region is used to force a selection or a deselection and therefore moving the central region over a portion of the image causes that portion of the image to be selected or deselected, respectively. The peripheral region of the cursor surrounds the central region and defines an area where a hybrid level set algorithm for both boundary detection and region definition, particularly a matting region, is performed. This provides highly accurate boundary detection and matting region selection within a narrowly-focused peripheral region and eliminates the need to subsequently designate a matting region and apply a matting algorithm to complex portions of an object selection.

Abstract: Techniques are disclosed for color selection in a desktop publishing application. A color selection technique includes receiving a selection of an image, automatically sampling a color from a pixel of the selected image, and adding the sampled color to a color swatch in a graphical user interface. The sampled color may be the predominant color in the image (e.g., the color that appears in the greatest number of pixels), or the sampled color may be the darkest or lightest color in the image. In another embodiment, several colors (e.g., two, three, four, five, six, seven, eight, nine or ten) are sampled from different pixels of the selected image, and some or all of the sampled colors are added to the color swatch. A designer can then select the sampled color(s) from the color swatch and apply the selected color(s) to one or more elements of a layout.

Abstract: Embodiments of the present invention facilitate automatically white-balancing images captured under mixed lighting conditions. More particularly, some embodiments are directed to leveraging flash photography to capture two images in quick succession, one with the flash activated and one without. By combining these two images, a per-pixel white balance kernel can be automatically determined and used to generate a white-balanced image without requiring any user input or user assumptions about the lighting in the scene. In embodiments, the white balance kernels for pixels that lie in the shadow or appear as specular spots are masked and can be interpolated from the white balance kernels at unmasked kernels. Motion effects between the flash and non-flash images may also be reduced.

Abstract: One exemplary embodiment involves receiving, from an administrator device, a first request to provide a set of privileges to at least one of a set of users and a set of user groups to access private resources referenced by a path of a networked hierarchical repository, the set of privileges including at least one authorized access privilege. The embodiment also involves modifying a metadata associated with the private resources to indicate the set of privileges to provide to the at least one of the set of users and the set of user groups, the metadata associated with the private resources being modified to at least indicate the authorized access privilege. Additionally, the embodiment involves receiving a second request for authorized access to a set of resources associated with the networked hierarchical repository and providing the authorized access based on whether the metadata associated with the set of resources indicates to provide the authorized access.

Abstract: Certain embodiments of the present disclosure provide methods and systems for detecting unsolicited messages (i.e., spam) and preventing unsolicited messages from being sent. Certain embodiments provide analyzing dynamic information received as part of a request to generate a message. Parts or elements of the dynamic information such as sender, recipient and content are analyzed using filters and configurable constants to determine an abuse score indicative of the likelihood that the dynamic information may generate spam if included in an electronic message. A course of action may be followed dependent on the abuse score determined. For example, if the dynamic information is deemed not likely to generate spam, an electronic message from the information received is generated and sent. Certain embodiments allow implementation of the system and method in a distributed environment comprising of multiple systems or servers performing the same method.

Abstract: Disclosed are various embodiments for performing color blending of document objects using GPU hardware of a computer. A computer obtains a document specifying a first object and a second object presented in the document, where a portion of the second object overlaps the first object. The computer renders, on a display device, a presentation of the document, where the rendered presentation including the first object of the document. In response to determining that the document specifies blending the second object with the first object in the overlapping portion, configuring dual-source color circuitry of a graphics processing unit (GPU) of the computer to compute, via a dual-source color blending operation, a result color for the blended first and second objects. The computer renders, on the display device, the second object of the document using the result color for the overlapping portion.

Abstract: Techniques disclosed herein identify visitors who visit a real-world shopping venue as belonging to a group. This involves tracking a physical action of a visitor present at the real-world shopping venue and identifying the visitor based on the tracked physical action and previously collected visitor information associated with the venue. The visitor is determined to be a member of a group based on the previously collected visitor information. Based on determining that the visitor is a member of the group, a notification is sent to another member of the group. The notification identifies that the visitor has placed at least one item in a sharing cart while the visitor is present at the real-world shopping venue. A system determines that visitors who routinely visit a real-world shopping venue belong to a group based on their activities while visiting the venue, and allows the visitors to provide feedback to each other.

Abstract: Residual vectors are compressed in a lossless compression scheme suitable for cloud DVR video content applications. Thus, a cloud DVR service provider can take many copies of the same file stored in the cloud and save storage space by compressing those copies while still maintaining their status as distinct copies, one per user. Vector quantization is used for compressing already-compressed video streams (e.g., MPEG streams). As vector quantization is a lossy compression scheme, the residual vector has to be stored to regenerate the original video stream at the decoding (playback) node. Entropy coding schemes like Arithmetic or Huffman coding can be used to compress the residual vectors. Additional strategies can be implemented to further optimize this residual compression. In some embodiments, the techniques operate to provide a 25-50% improvement in compression. Storage space is thus more efficiently used and video transmission may be faster in some cases.

Abstract: Techniques are disclosed for performing flood-fill operations on vector artwork. In one embodiment, a region under a point of interest (POI) of vector artwork is rasterized and flood-filled, and an initial bounding shape around that area is used as a first guess as to the area to be filled. In other cases, the initial bounding shape is created around some initial area that includes the POI (no rasterization). In any such case, vector objects having bounding shapes that intersect the initial bounding shape are identified and fed into a planar map. After map planarization, a new bounding shape is created around a new area resulting from the planarizing and that includes the POI. In response to that bounding shape not extending beyond the initial bounding shape, a vector-based flood-fill operation can be performed on that new area. The process repeats if a new bounding shape extends beyond previous bounding shape.