A system, method and a computing device for performing a method for data augmentation allowing for document classification of a plurality of documents are disclosed. The system, method and computing device including a processor configured to convert the plurality of documents into images, a memory configured to store the images, the processor configured to obtain a vector representation for each page included in the plurality of documents, the processor configured to create a plurality of clusters from the images based on similarity, where each cluster of the plurality of clusters represents a distinct page format, the processor configured to select one image from each cluster of the plurality of clusters, the processor configured to compile the selected one image from each cluster of the plurality of clusters to create a logically complete document, the memory configured to store the logically complete document, and the processor configured to train the classification based on the complete document.

There are provided a font creation apparatus, a font creation method, and a font creation program capable of generating, from small-number character images having a desired-to-be-imitated style, a complete font set for any language having the same style as the character images. A feature amount extraction unit (40) receives a character image (32) of a first font having a desired-to-be-imitated style and extracts a first feature amount of the first font of the character image (32). An estimation unit (42) estimates a transformation parameter between the extracted first feature amount and a second feature amount of a reference second font (34). A feature amount generation unit (44) generates a fourth feature amount of a second font set to be created by transforming a third feature amount of a complete reference font set (36) based on the estimated transformation parameter. A font generation unit (46) generates a complete second font set by converting the generated fourth feature amount of the second font set into an image.

A computer transforms high-dimensional data into low-dimensional data. (A) A distance matrix is computed from observation vectors. (B) A kernel matrix is computed from the distance matrix using a bandwidth value. (C) The kernel matrix is decomposed using an eigen decomposition to define eigenvalues. (D) A predefined number of largest eigenvalues are selected from the eigenvalues. (E) The selected largest eigenvalues are summed. (F) A next bandwidth value is computed based on the summed eigenvalues. (A) through (F) are repeated with the next bandwidth value until a stop criterion is satisfied. Each observation vector of the observation vectors is transformed into a second space using a kernel principal component analysis with the next bandwidth value and the kernel matrix. The second space has a dimension defined by the predefined number of first eigenvalues. Each transformed observation vector is output.

This disclosure relates to various implementations that dynamically adjust one or more shallow depth of field (SDOF) parameters based on a designated, artificial aperture value. The implementations obtain a designated, artificial aperture value that modifies an initial aperture value for an image frame. The designated, artificial aperture value generates a determined amount of synthetically-produced blur within the image frame. The implementations determine an aperture adjustment factor based on the designated, artificial aperture value in relation to a default so-called “tuning aperture value” (for which the camera's operations may have been optimized). The implementations may then modify, based on the aperture adjustment factor, one or more SDOF parameters for an SDOF operation, which may, e.g., be configured to render a determined amount of synthetic bokeh within the image frame. In response the modified SDOF parameters, the implementations may render an updated image frame that corresponds to the designated, artificial aperture value.

A real-time end-to-end system for capturing ink strokes written with ordinary pen and paper using a commodity video camera is described. Compare to traditional camera-based approaches, which typically separate out the pen tip localization and pen up/down motion detection, described is a unified approach that integrates these two steps using a deep neural network. Furthermore, the described system does not require manual initialization to locate the pen tip. A preliminary evaluation demonstrates the effectiveness of the described system on handwriting recognition for English and Japanese phrases.

The information processing device generates headings from structured documents. The acquisition means acquires a structured document including headings and texts. The feature word extraction means extracts feature words from subordinate elements of the heading for the headings included in the structured document. The heading generation means generates a new heading corresponding to the subordinate elements based on the extracted feature words.

The method of matching digital images of the same article in a data processor unit comprises the steps of: transforming each digital image of an article into a local divergence topographic map of the luminance gradient vector field; detecting singularities or extrema of local divergence in the luminance gradient vector field, such singularities corresponding to points of interest in said digital image; and, for each detected point of interest, encoding the values for the singularity of the gradient field that are located on a plurality of concentric rings centered on the point of interest so as to derive a digital data vector (210); and transforming said vector into a digital hash key (220) by means of a family of hash functions of the cosine Locality-Sensitive Hashing (LSH) type.

An image processing method implemented by a computer includes extracting feature points from captured images that are sequentially generated by an image capture device and include at least a first captured image and a second captured image generated prior to the first captured image, determining whether the number of feature points extracted from the first captured image exceeds a threshold value, and specifying a location of the first captured image relative to the second captured image upon determining that the number of the feature points extracted from the first captured image is below the threshold value.

A computer-implemented method, includes accessing, by a processor, a set of asset documents. The method also includes performing, by the processor, feature extraction on text of each document of the set of asset documents using a feature extraction module to generate a set of features, where each feature of the set of features represents a document of the set of asset documents. The method also includes generating, by the processor, a set of lower-dimensional features from the set of features using a singular value decomposition module. The method also includes generating, by the processor, a set of clusters from the set of lower-dimensional features using a clustering module. The method also includes training, by the processor, a machine-learning model of a classification microservice using the set of clusters generated from the clustering module.

Implementations of the present specification disclose image recognition methods, apparatuses, and devices, and authentication methods, apparatuses, and devices. A solution includes the following: obtaining a target image of a target object, where before the target image is obtained, a recognition identifier is mapped onto the target object, and where the recognition identifier is used to form a corresponding recognition feature in the target image; and determining an attribute of the target object corresponding to the target image based on a predetermined mapping relationship and the recognition feature, where the mapping relationship includes a corresponding relationship between the recognition feature and the attribute of the target object.