Abstract: Methods of detecting an object in an image using a convolutional neural-network-based architecture that processes multiple feature maps of differing scales from differing convolution layers within a convolutional network to create a regional-proposal bounding box. The bounding box is projected back to the feature maps of the individual convolution layers to obtain a set of regions of interest (ROIs) and a corresponding set of context regions that provide additional context for the ROIs. These ROIs and context regions are processed to create a confidence score representing a confidence that the object detected in the bounding box is the desired object. These processes allow the method to utilize deep features encoded in both the global and the local representation for object regions, allowing the method to robustly deal with challenges in the problem of object detection. Software for executing the disclosed methods within an object-detection system is also disclosed.

Abstract: An imaging system comprising: a pair of cameras (2, 3) adapted to capture images in the near infrared region of the electromagnetic spectrum; a camera (4) adapted to capture images in the visible region of the electromagnetic spectrum; a near-infrared light source (5); and a controller (6). The system configured to obtain a plurality images of the face of an individual and, from a local binary pattern distribution of image pixels, determine the age of the individual in each of the images by applying a linear regression technique and computing an average.

Abstract: A method to characterize shape variations in brain ventricles during embryonic growth in mammals, the method including extracting a brain ventricle skeleton from one or more images, calculating a volume profile for the skeleton using the extracted images, partitioning the brain ventricle based on the volume profile along the skeleton, the brain ventricle being partitioned into two lateral ventricles and a main ventricle, the main ventricle being further partitioned into three sub regions, determining volume vectors of the two lateral ventricles and the three sub regions, computing a means square error between the determined computed volume vectors and a pretrained mean volume vector of embryos during different gestational stages, and classifying the embryo to the gestational stage having the lowest mean square error.

Abstract: A method and system for authenticating users accessing financial accounts from user terminals via multi-factor image authentication. The system includes an authentication server and a user terminal. The method captures a first facial image of a user and compares this image with stored facial recognition credentials. The method prompts the user via a facial gesture cue to make a facial gesture, captures a second facial image of the user, and compares the second image with stored facial gesture credentials. The user is authorized to perform a transaction in the event the first facial image matches a facial recognition credential for an authorized account, and the second facial image matches a facial gesture credential associated with the authorized account. Facial gesture credentials may be based upon static gestures or dynamic gestures, and may be overt or secret. An additional authentication factor may employ a secondary biometric sensor.

Abstract: A method and system for authenticating users accessing financial accounts from user terminals via multi-factor image authentication. The system includes an authentication server and a user terminal. The method captures a first facial image of a user and compares this image with stored facial recognition credentials. The method prompts the user via a facial gesture cue to make a facial gesture, captures a second facial image of the user, and compares the second image with stored facial gesture credentials. The user is authorized to perform a transaction in the event the first facial image matches a facial recognition credential for an authorized account, and the second facial image matches a facial gesture credential associated with the authorized account. Facial gesture credentials may be based upon static gestures or dynamic gestures, and may be overt or secret. An additional authentication factor may employ a secondary biometric sensor.

Abstract: Techniques (300, 400, 500) and apparatuses (100, 200, 700) for recognizing accented speech are described. In some embodiments, an accent module recognizes accented speech using an accent library based on device data, uses different speech recognition correction levels based on an application field into which recognized words are set to be provided, or updates an accent library based on corrections made to incorrectly recognized speech.

Abstract: A computer-implementable method for determining a score for matching images includes extracting and determining. Features are extracted from each of a query image acquired in a first modality and a database image acquired in a second modality. A vector distance distribution is determined based on a codebook for the first modality for each of the query image and the database image represented by the extracted features. The method determines a set of distances between the vector distance distribution determined for the query image and the vector distance distribution determined for the database image. The score for matching the query image and the database image is determined based on the determined set of distances.

Abstract: A traffic light recognition device includes: an imaging attitude setting unit configured to set an imaging direction of the camera on the basis of a position on the image of the traffic light and of a moving direction in the future on the image of the traffic light; a camera attitude controller configured to change the imaging direction of the camera to an imaging direction set by an imaging direction setting unit; and a traffic light recognition unit configured to recognize the traffic light from an image captured in the imaging direction by the camera. The traffic light recognition device is capable of accurately detecting the lighting state of a traffic light without being affected by the attitude of a camera.

Abstract: An image compression method includes compressing an input image with first and second compression methods to generate first and second (e.g., lossless and lossy) compressed images. First and second residual layers are generated, based on a difference between the first and second compressed images. Connected components in the residual layers are identified. Each connected component includes a group of one or more pixels that, when mapped to the second compressed image is connected, in first and second directions, to pixels in the second compressed image. A compressed image is generated, which includes, for a connected component identified in the first residual layer, removing corresponding pixels from the second compressed image, and for a connected component identified in the second residual layer, adding corresponding pixels to the second compressed image. The system and method can thus provide file size savings associated with lossy compression while avoiding character replacement.

Abstract: A signal processing apparatus and a related method. In particular, the apparatus includes a novel image reconstructor to reconstruct cross sectional images of a specimen from interferometric projection data (m). The reconstructor (RECON) is based on a new forward model that accounts for cross-talk of a phase contrast signal into a dark field signal.

Abstract: An electronic device and a voice recognition method thereof are provided. A voice recognition method of an electronic device includes receiving trigger voice, storing a characteristic of the trigger voice for voice verification by analyzing the trigger voice. When a user issues a voice command, determining whether the user voice command comprises the characteristic belonging to the trigger voice, and performing function of the electronic device corresponding to the user voice command based on the determination. Accordingly, users may easily use the voice recognition function of the electronic device.

Abstract: An image recognizing apparatus includes a processor that controls first and second learning processes, the first learning process in second layers including holding, based on a large/small relation between neuron data size and parameter size of the second layer, in a memory area, an error gradient of parameters to be sent to the corresponding layer of the second layers; and the second learning process between first layers including holding, in a memory area of each first layers, an error gradient of parameters to be sent to the corresponding layer of the first layers, which is computed based on the error gradient or an error gradient of a previous layer of the first layers, based on a large/small relation between neuron data size and parameter size of the first layer.

Abstract: Methods and systems for monitoring a video feed include capturing input data representing a monitored area using one or more cameras to produce an input data set comprising individual images. Initial segmentation scores are determined for each image of the input data set using a neural network, with each image being assigned an initial segmentation score for each of multiple segmentation classes. Final segmentation scores are determined for each image of the input data set by enforcing a smoothness criterion. The input data set is segmented in accordance with the final segmentation scores. It is determined whether an alert condition is met based on segmented data set. An alert is generated if the alert condition is met.

Abstract: A method for compositing images by an electronic device is described. The method includes obtaining a first composite image that is based on a first image from a first lens with a first focal length and a second image from a second lens with a different second focal length. The method also includes downsampling the first composite image to produce a downsampled first composite image. The method further includes downsampling the first image to produce a downsampled first image. The method additionally includes producing a reduced detail blended image based on the downsampled first composite image and the downsampled first image. The method also includes producing an upsampled image based on the reduced detail blended image and the downsampled first composite image. The method further includes adding detail from the first composite image to the upsampled image to produce a second composite image.

Abstract: A method and system for authenticating a user. An expected digital representation of an iris of at least one authentic user is a combination of a first digital representation of an iris of an eye of at least one authentic user and a second digital representation of at least one authentic overlay pattern. In response to a determination that a third digital representation of a captured image of an iris of a user to be authenticated does not match the expected digital representation: the expected digital representation is modified to alter a relative rotational alignment between the representation of the at least one authentic overlay pattern and the expected digital representation of the iris; and a further comparison between the modified expected digital representation and the third digital representation is performed. From the further comparison, an authentication signal indicative of whether authentication of the user has been successful is generated.

Abstract: A pattern classifying apparatus according to the present invention accesses at least two of a plurality of memories holding data to be used in pattern classification in parallel so as to read a plurality of data pieces and performs pattern classification with weak classifiers by using the plurality of read data pieces.

Abstract: Methods and an apparatus for background subtraction are described. According to some implementations, the temporal history of a pixel location can be used to classify whether a current input pixel of a current input video frame is a background pixel or a foreground pixel. The classification can be made based on the number of matching pixels in corresponding pixel locations in background history planes. One of the background history planes can be updated according to the classified background pixels and persistent classified foreground pixels.

Abstract: A method of speeding up image detection, adapted to increase a speed of detecting a target image and enhance efficiency of image detection, comprises the steps of capturing an image; retrieving a plurality of characteristic points of the image; creating a region of interest (ROI) centered at the characteristic points each; creating a plurality of search point scan windows corresponding to the ROIs, respectively; calculating target hit scores of the characteristic points and the search point scan windows; comparing the target hit scores of the characteristic points and the search point scan windows to obtain an ROI most likely to have a target image; calculating centroid coordinates of the ROI by a centroid shift weight equation; and narrowing a scope of ROI search according to a location of the centroid coordinates and reducing a displacement between the search points.

Abstract: An unorganized point cloud may be created by an optical 3D scanner that scans a physical object, or by computer simulation. The point cloud may be converted into binary raster layers, which encode material deposition instructions for a multi-material 3D printer. In many cases, this conversion—from point cloud to binary raster files—is achieved without producing a 3D voxel representation and without producing a boundary representation of the object to be printed. The conversion may involve spatial queries to find nearby points, filtering material properties of the found points, looking up material mixing ratios, and dithering to produce binary raster files. These raster files may be sent to a multi-material 3D printer to control fabrication of an object. A user interface may display a preview of the object to be printed, and may accept user input to create or modify a point cloud.

Abstract: A fingerprint registration method, device and terminal thereof are proposed. The method includes: retrieving feature information of a fingerprint recoded every time based on preset index types when registering the fingerprint; matching the feature information based on preset index information corresponding to the index types to determine whether the fingerprint recorded every time is effective; and signaling a successful fingerprint registration if a number of effective fingerprints recorded is monitored to reach a preset number of fingerprint templates.