Abstract: A shape of an object is represented by a set of points inside and outside the shape. A decision function is learned from the set of points an object. Feature points in the set of points are selected using the decision function, or a gradient of the decision function, and then a local descriptor is determined for each feature point.

Abstract: In one embodiment, there is provided a method for an Optical Character Recognition (OCR) system. The method comprises: recognizing an input character based on a plurality of classifiers, wherein each classifier generates an output by comparing the input character with a plurality of trained patterns; grouping the plurality of classifiers based on a classifier grouping criterion; and combining the output of each of the plurality of classifiers based on the grouping.

Abstract: Techniques and structures are disclosed in which one or more distortion categories are identified for an image or video, and a quality of the image or video is determined based on the one or more distortion categories. The image or video may be of a natural scene, and may be of unknown provenance. Identifying a distortion category and/or determining a quality may be performed without any corresponding reference (e.g., undistorted) image or video. Identifying a distortion category may be performed using a distortion classifier. Quality may be determined with respect to a plurality of human opinion scores that correspond to a particular distortion category to which an image or video of unknown provenance is identified as belonging. Various statistical methods may be used in performing said identifying and said determining, including use of generalized Gaussian distribution density models and natural scene statistics.

Abstract: Method and apparatus for summarizing video data is disclosed. In one embodiment, a method includes accessing face information that is associated with at least one identified person within frames of the video data, examining user specifications for selecting portions of the video data, comparing the user specifications with the face information to determine indicia of interest related to each of the frames, identifying at least one of the frames that are in accordance with user specifications based on the indicia of interest and forming summary video data using the at least one identified frame of the video data.

Abstract: A classifier training system trains unified classifiers for categorizing videos representing different categories of a category graph. The unified classifiers unify the outputs of a number of separate initial classifiers trained from disparate subsets of a training set of media items. The training process divides the training set into a number of bags, and applies a boosting algorithm to the bags, thus enhancing the accuracy of the unified classifiers.

Abstract: Embodiments include a scene classification system and method. In one embodiment, a method includes forming a first plurality of image features from an input image, processing the first plurality of image features in the first scene classifier.

Abstract: Various embodiments of the invention provide systems and methods for extracting information from digital documents, including physical documents that have been converted to digital documents. For example, some embodiments are configured to extract information from a field in a digital document by identifying a block of tokens before (i.e., a prior block) and a block of tokens after (i.e., a post block) the field from which the information is to be extracted, where both the prior block and post block are known to be associated with the field type of the field (e.g., name, address, phone number, etc.).

Abstract: A display device and a control method thereof are provided. The display device includes a camera obtaining an image and a controller obtaining the direction of a user included in the obtained image and correcting the image such that the direction of the user is synchronized with the photographing direction of the camera. Even when the direction of the user does not correspond to the photographing direction of the camera, an image of the user can be corrected to correctly recognize a user's gesture.

Abstract: An information processing apparatus includes: a distinguishing unit which, by using an ensemble classifier, which includes a plurality of weak classifiers outputting weak hypotheses which indicates whether a predetermined subject is shown in an image in response to inputs of a plurality of features extracted from the image, and a plurality of features extracted from an input image, sequentially integrates the weak hypotheses output by the weak classifiers in regard to the plurality of features and distinguishes whether the predetermined subject is shown in the input image based on the integrated value. The weak classifier classifies each of the plurality of features to one of three or more sub-divisions based on threshold values, calculates sum divisions of the sub-divisions of the plurality of features as whole divisions into which the plurality of features is classified, and outputs, as the weak hypothesis, a reliability degree of the whole divisions.

Abstract: A personal authentication system according to the present invention includes a matrix generation unit, a feature extraction unit, a feature transformation unit, a processing unit, and a data matching unit. The matrix generation unit classifies a plurality of feature amounts of face image data, which are recorded in advance, into classes and calculates as a mapping matrix, coefficients of a linear discriminant equation which uses the recorded plurality of feature amounts as variables. The feature extraction unit extracts a first feature amount from first face image data and extracts a second feature amount from second face image data. The feature transformation unit, by using the mapping matrix, transforms the first feature amount into a first transformed feature amount and transforms the second feature amount into a second transformed feature amount.

Abstract: Techniques and methods are disclosed herein for combining and weighting of values from and associated with classifiers. Classifiers are used to recognize characters as part of an optical character recognition (OCR) system. Various methods of normalization facilitate combining of results of classifiers. For example, weight values may be entered into a weight table having two columns, one that includes weights from comparing patterns with images of correct characters, the other column includes weights from comparing patterns with images of incorrect characters.

Abstract: An image processing apparatus determines a specific color region that includes a specific color, and a non-specific color region around which the specific color regions are present in three directions, i.e., an upper, left, and right directions, in a picture in the non-specific color regions that are not determined as including the specific color as specific regions to be subjected to predetermined image processing.

Abstract: Provided are an image processing method and apparatus for effectively reducing noise in an image, and a recording medium storing a program for executing the method. The image processing apparatus includes a noise reduction unit configured to apply a noise reduction filter to first image data to obtain second image data; a first edge data obtaining unit configured to calculate edge data lost in the second image data compared with the first image data to obtain the first edge data; and a first synthesis unit configured to tune the first edge data and calculate third image data from the second image data and the tuned first edge data.

Abstract: A system and method for adjusting perspective and disparity in a stereoscopic image pair using range information includes receiving the stereoscopic image pair representing a scene; identifying range information associated with the stereoscopic image pair and including distances of pixels in the scene from a reference location; generating a cluster map based at least upon an analysis of the range information and the stereoscopic images, the cluster map grouping pixels of the stereoscopic images by their distances from a viewpoint; identifying objects and background in the stereoscopic images based at least upon an analysis of the cluster map and the stereoscopic images; generating a new stereoscopic image pair at least by adjusting perspective and disparity of the object and the background in the stereoscopic image pair, the adjusting occurring based at least upon an analysis of the range information; and storing the new generated stereoscopic image pair in a processor-accessible memory system.

Abstract: A wordspotting system and method are disclosed. The method includes receiving a keyword and, for each of a set of typographical fonts, synthesizing a word image based on the keyword. A keyword model is trained based on the synthesized word images and the respective weights for each of the set of typographical fonts. Using the trained keyword model, handwritten word images of a collection of handwritten word images which match the keyword are identified. The weights allow a large set of fonts to be considered, with the weights indicating the relative relevance of each font for modeling a set of handwritten word images.

Abstract: According to one embodiment, an electronic apparatus includes a storage device which stores face thumbnail indexing information including face images and time stamp information, extracting module configure to assign time zones to a video content data and to extract face images belonging to each time zone based on a time stamp information, classifying module configure to classify facial images of the same person from the extracted facial images, calculating module configure to calculate a frequency of appearance of each classified facial image, and facial image indication module configure to display a list of the facial images included in the facial image indexing information in a facial image indication in a two-dimensional display area, the facial image indication having time-zone-specific display areas in columns corresponding to the time zones, each facial image displayed in each time-zone-specific display area being displayed in a size based on the frequency of appearance.

Abstract: In one embodiment, L dimensional images are trained, mapped, and aligned to an M dimensional topology to obtain azimuthal angles. The aligned L dimensional images are then trained and mapped to an N dimensional topology to obtain 2N vertex classifications. The azimuthal angles and the 2N vertex classifications are used to map L dimensional images into O dimensional images.

Abstract: An object of the invention is to provide a material determining apparatus for a steel product and a material determining method for a steel product which can stably and precisely determine a carbon content of the steel product.

Abstract: A computer program and apparatus function to extract objects from a video by identifying regions to be tracked, tracking the regions across several frames, calculating motions of the regions, and identifying new regions to be tracked. Regions of a selected frame of the video that are appropriate for tracking are selected. Tracking of the identified regions is then performed across frames subsequent to the selected frame of the video. Motions of the identified regions are calculated, and new regions for tracking are identified.

Abstract: The present invention relates to a lane-dividing line detection device and has an object of detecting a lane-dividing line with high accuracy by accurately extracting characteristic points of the lane-dividing line while responding flexibly to road surface conditions. According to the present invention, pixel parts where a brightness variation is larger than a predetermined threshold are extracted from an image picked up by a camera that picks up an area ahead of a vehicle as edge points representing the lane-dividing line on a road surface (step 106). Next, candidates for the lane-dividing line drawn on a road are generated based on the extracted edge points (step 108). Then, the predetermined threshold for the brightness variation used to extract the edge points is changed based on the number of the generated candidates for the lane-dividing line (steps 122 through 130).