Abstract: A computer-implemented technique is described for generating a textual narrative based on a set of input images. In one scenario, the end user captures the set of input images while visiting one or more locations. The generated narrative describes the user's travel experience at those locations in a cohesive manner. In one implementation, the technique generates the narrative based on information extracted in offline fashion by a knowledge acquisition component. The knowledge acquisition component, in turn, produces the knowledgebase by mining one or more knowledge sources (such as one or more travel blogs) that provide image-annotated textual travel narratives. The technique can therefore be said to reuse common elements in the travel narratives of others in describing the user's own set of input images.

Abstract: Systems and methods for managing and processing building plan documents. An electronic building plan document including a plurality of plan sheets is received. A first plan sheet is provided for display. A user interface is provided via which a user can select or enter a first comment associated with the first plan sheet. A user interface is provided via which the user can associate metadata, including a project type or discipline, with the first comment. A first plurality of comments, including the first comment, is stored in association with respective metadata. A user interface via which the user can select, by category and/or subcategory a plurality of comments to be included in a plan correction list, and a correction list is generated including a plurality of comments specified by a plurality of users.

Abstract: Techniques for rule creation from natural language text (NLT) are disclosed. In an embodiment, a rule statement in the NLT is grammatically parsed to obtain a parse tree. Further, each of noun phrase (NP) core sub-tree and verb phrase (VP) core sub-tree, in the parse tree, is partitioned into at least one sub-tree. Furthermore, one or more operators are extracted using at least one sub-tree of NP or VP core sub-trees. The rule statement is substantially simultaneously splitted into groups of 3 adjacent words (3-grams). Each of the 3-grams is then compared with a predefined list to extract data properties and class concepts associated with the rule statement. Moreover, subscripts are assigned to the data properties, class concepts and operators. A rule head is then created based on the data properties and operators. Also, a rule is created using the data properties, operators, class concepts, associated subscripts and rule head.

Abstract: A method of analyzing text is performed at an electronic system that includes one or more processors and memory storing instructions for execution by the one or more processors. The method includes extracting terms from a corpus of text and generating a plurality of rankings of the extracted terms. Each ranking of the plurality of rankings is based on a respective metric of a plurality of metrics. The method further includes defining a set of terms, which includes selecting a number of top-ranked terms from each ranking of the plurality of rankings. The set of terms may be provided for display and/or used as search terms to perform a search.

Abstract: The invention discloses a method and device for recognizing a domain named entity. The main technical solution is: conducting word segmentation on a text to be recognized; tagging each segmented word in the text to be recognized according to a tag set of a domain corresponding to the text to be recognized, where the tag set includes a basic tag set based on domain named entity recognition and a tag set belongs to the corresponding domain, and the basic tag set includes a location tag of related words which constitute the domain named entity; and extracting the tagged segmented words according to a domain named entity extraction rule and constituting the domain named entity.

Abstract: A system and method for lowering probability that low data rates, high bit error rates, and susceptibility to signal degradation in communication environments during storms, solar activity, and adversarial activity is disclosed. The system and method affect communications by minimizing the amount of data/information that needs to be transmitted, partly by creating an algorithmic process/method that moves knowledge and not data and information. The system and method is based upon the realization that structured communications often possess a similar level of context that can be exploited to communicate full meaning (knowledge), even when only a small fraction of the message is transmitted to the receiver. Reducing the number of bytes transmitted significantly reduces the probability that a transmission will be affected by either naturally occurring or human supplied factors present in modern communication environments.

Abstract: A method and device for automatically changing or updating a configuration or setting of a communication system is disclosed. In one aspect, the method includes providing information to the communication system, the information comprising natural human language, storing the information in a digital storage device, detecting a triggering event in the information, and changing the configuration or setting of the communication system automatically using a processor. The information is an input to the communication system, an input from at least one alternate communication system, or a combination of an input to the communication system and an input from the at least one alternate communication system.

Abstract: A bilingual-corpus generating method includes: acquiring a first clause in the first language, a first translated clause obtained by translating the first clause into a second language, a second clause in the first language, a second translated clause obtained by translating the second clause into the second language; and generating, as a translated clause of a third clause obtained by constructing the first clause and the second clause as one sentence by using a first conjunction in the first language, a third translated clause obtained by constructing the first translated clause and the second translated clause as one sentence by using a second conjunction in the second language.

Abstract: A communication assistive device and a communication assistive method for people with language disorder are provided. The communication assistive method comprises the steps of: receiving user touch screen input selecting one of images representing sentence components displayed on a screen, the sentence components such as words and postpositions constituting a sentence to be outputted; combining the received sentence components to generate a sentence; and outputting the generated sentence by voice, wherein input for at least one of the sentence components is performed by selection of an image coding a shape or meaning of a character, wherein a widget is created for the generated sentence.

Abstract: A magnetic information processing apparatus for registering transaction information on a line in a document and storing the location of the line in magnetic media attached to the document, includes a magnetic head, a document receptacle configured to receive the document on which the magnetic media storing the position where the immediately previous transaction information is registered is attached, and a processing unit. The processing unit is programmed to control the magnetic head to read information from a first position of the magnetic media, move to a second position of the magnetic media, and read additional information from the second position of the magnetic media, extract registration position information indicating the location of the most recent previous registration position from the result of reading of the first and second positions of the magnetic media by the magnetic head, and output the extracted registration position information.

Abstract: The present invention provides a method and an apparatus for detecting anti-skimming magnetic interference, wherein the method comprises the steps of: S1: generating an original magnetic strip signal; acquiring an interfered magnetic strip signal by combining the original magnetic strip signal and an interference signal of an electromagnetic interference source to be tested; S2: decoding the interfered magnetic strip signal to obtain a decoded parameter; and S3: determining whether an original magnetic strip parameter is consistent with the decoded parameter; if so, obtaining a first test result that the electromagnetic interference source is unqualified to satisfy a magnetic card reading device's requirement for electromagnetic interference; otherwise, obtaining a second test result that the electromagnetic interference source is qualified to satisfy the magnetic card reading device's requirement for electromagnetic interference whereby security of the magnetic card reading device can be enhanced and the mag

Abstract: Various switchable RFID devices are disclosed. These switchable RFID devices may include one or more RFID tags and one or more switches. Some of these one or more switches are optionally wireless. In various embodiments, the switchable RFID devices include cellular phones, security devices, identity devices, financial devices, remote controls, and the like. The switchable RFID devices are optionally disposed in a passport. The passport may include radio frequency shielding.

Abstract: Systems and methods to determine motion parameters of physical objects using radio frequency identification (RFID) tags attached to the objects. In one embodiment, a method implemented in a radio frequency identification (RFID) system includes determining a motion parameter of the RFID tag based on detecting a Doppler frequency shift in a radio frequency signal received from the RFID tag.

Abstract: An Integrated Circuit (IC) for an RFID tag includes at least two antenna ports for coupling to at least two antennas. The IC may be configured to determine the port from which it receives an input signal, and provide a first functionality if it receives the input from a first port and a second functionality if it receives the input from a second port. The IC may be configured to determine and/or offer a functionality based on the receiving port.

Abstract: An authentication method includes RFID tags authenticating RFID readers. A tag sends a tag identifier and a reader challenge to a reader in response to one or more commands from the reader. The reader then either derives a response to the reader challenge itself or has a verification authority derive the response. The response may be derived from parameter(s) in the reader challenge, and may be derived using a cryptographic key. The reader then sends the response to the tag along with one or more commands. The tag verifies the response before executing action(s) associated with the command(s).

Abstract: Tamper detection on an article is performed by reading data from an RFID tag having an antenna with segments designed to break apart. The RFID tag is secured to the article in such a way that, when the article is tampered with, the antenna segments will break apart. Detachment would decrease the reading range of the RFID tag. An RFID reader may be used to read the data from the RFID tag. The predefined reading range (Rp) of the RFID tag, with antenna segments entact, is obtained from the data which was read by the RFID reader. The RFID reader obtains the data while located at an actual reading range (Ra) relative to the RFID tag. By comparing Ra with Rp, it is determined whether the antenna is still entact. If Ra<Rp, the antenna is likely to be broken apart, thereby indicating that the article has been tampered with.

Abstract: A barcode detection method includes acquiring an image by a camera, acquiring a horizontal gradient and a vertical gradient of each pixel of the image within a region, generating a gradient phase and gradient magnitude of each pixel according to the horizontal gradient and the vertical gradient, performing a binarization process to the gradient magnitude of each pixel of the image within the region for generating a binarized image, vertically and horizontally sampling the binarized image for generating the gray level flip count in a vertical direction and a horizontal direction, locating an image region of a barcode according to the gray level flip count in the vertical direction and the horizontal direction, acquiring a plurality of gradient phases of all pixels within the image region for generating a gradient phase distribution, and generating a barcode format detection result according to the gradient phase distribution.

Abstract: Various embodiments described herein relates to indicia reading system for scanning and decoding direct part markings (DPM) indicia. The indicia reading system includes a plurality of collimating optical elements that generate collimated light beams. A plurality of diffractive optical elements is tuned to split a corresponding collimated light beam into at least two sets of refracted light beams at two incident angles relative to a DPM indicia. A first set of refracted light beams with a zero-diffraction order and a first incident angle causes a bright-field illumination in a far-field area. A second set of refracted light beams with a first-diffraction order and a second incident angle causes a dark-field illumination in a near-field area. An image sensor generates image data of the DPM indicia based on a set of reflected light beams received from the DPM indicia in accordance with one of the bright-field illumination or the dark-field illumination.

Abstract: In one aspect of the invention for which protection is sought there is provided reader apparatus for reading an identification code carried by a surface feature device, the surface feature device having a plurality of respective surface feature portions each provided at respective lateral locations over a lateral surface at one of a plurality of predetermined distances from a lateral reference plane, the apparatus comprising a light source and a light detector, the apparatus being configured to direct light from the light source onto the respective surface feature portions of the surface feature device and to detect by means of the detector light scattered by the respective surface feature portions, the detector being configured to output a detector signal indicative of the intensity of light incident thereon, the apparatus being configured to generate a reader signal responsive to the distance of each of the respective surface feature portions of the surface feature device from the reference plane in depende

Abstract: Embodiments are directed to methods and systems to increase flow and throughput of particular data through the use of a machine readable code with portion analysis. The machine readable code may be generated by a transaction processing network and correspond with one or more resource provider computers. When data is received by the transaction processing network that corresponds with one of the resource provider computers, a portion of the machine readable code that is proportional to a size or measurement of the data may be provided in exchange for the data. The recipient of the portion of the code and, once the recipient collects a predetermined number of machine readable code portions to complete the machine readable code, the complete machine readable code may be scanned and transmitted back to the transaction processing network to be exchanged for additional information or to initiate further processing.

Abstract: A two-dimensional barcode encoder generates the hybrid two-dimensional barcode image that has control features corresponding to control features of a first formatting standard. The barcode encoder encodes a first data set in a data portion of the two-dimensional barcode in a second format that is different from a data encoding format of the first formatting standard. The encoder may also encode a second data set in the data portion of the two-dimensional barcode in the data encoding format of the first formatting standard. The encoder may also constellation encode the first data set.

Abstract: The present disclosures relates to finding or localizing machine readable indicia (e.g., a barcode or digital watermark) in imagery. One claim recites an apparatus comprising: memory for buffering blocks of image data, the image data having been captured with a camera and depicting a printed object; one or more processors programmed for: generating an edge orientation sensitive feature set from the image data; using a first trained classifier to determine whether the feature set includes data representing a barcode; and using N additional trained classifiers to determine an orientation angle associated with the barcode, wherein N comprises an integer greater than 3, and wherein the orientation angle is selected based on a probability metric. Of course, other claims and combinations are provided too.

Abstract: A filter assembly for authenticating a filter cartridge includes a filter housing and an optical scanner mounted to the filter housing. A controller uses the optical scanner to take one or more digital images of the filter cartridge and obtain digital coded data identifying the filter cartridge. Specifically, the controller locates a functional locator on the filter cartridge, determines a location of an identifying region on the filter cartridge, reads the digital coded data by interrogating the identifying region with the optical scanner, and determines that the filter cartridge is authentic based at least in part on the digital coded data.

Abstract: Systems and methods for establishing optimal reading conditions for a machine-readable symbol reader. A machine-readable symbol reader may selectively control reading conditions including lighting conditions (e.g., illumination pattern), focus, decoder library parameters (e.g., exposure time, gain), etc. Deep learning and optimization algorithms (e.g., greedy search algorithms) are used to autonomously learn an optimal set of reading parameters to be used for the reader in a particular application. A deep learning network (e.g., a convolutional neural network) may be used to locate machine-readable symbols in images captured by the reader, and greedy search algorithms may be used to determine a reading distance parameter and one or more illumination parameters during an autonomous learning phase of the reader.

Abstract: An optical identification method for sensing a physiological feature, includes: projecting light to a physiological portion for generating reflection light from the physiological portion; receiving the reflection light, to generate an image; generating slant pattern information according to the image; transforming the slant pattern information into a pattern identification matrix; and determining the physiological feature according to the pattern identification matrix.

Abstract: A fingerprint sensing device includes: a fingerprint sensor including scan lines and initialization lines extending in a first direction and arranged in a second direction, sensing lines extending in the second direction and arranged in the first direction, and sensing pixels connected to the scan lines, the initialization lines, and the sensing lines and disposed in a matrix configuration; a scan driver configured to apply an enable-level initialization signal to a-numbered sensing pixel rows while applying a disable-level initialization signal to b-numbered sensing pixel rows, and applying an enable-level scan signal to one sensing pixel row in the b-numbered sensing pixel rows; and a sensing driver configured to receive a sensing signal from the sensing pixel row to which the scan signal is applied through the sensing lines, wherein the a and b are positive numbers that are greater than 2.

Abstract: A fingerprint sensor is disclosed. The fingerprint sensor includes a sensor plate, a voltage source, a control circuit and a first conductive plate. The sensor plate detects a first capacitance in response to a touch event on the fingerprint sensor. The voltage source establishes a predetermined voltage difference between a first node and a second node. The control circuit, connected to the sensor plate via the first node, charges a first capacitor associated with the first capacitance during a charge period, and discharges the first capacitor during a discharge period. The first conductive plate, separated from the sensor plate, is coupled with the voltage source via the second node. During the charge period and the discharge period of the control circuit, a voltage difference between the sensor plate and the first conductive plate is maintained at the predetermined voltage difference.

Abstract: A fingerprint sensor and a terminal device are provided, including: a first drive circuit, configured to generate a periodic driving signal according to a periodic first input signal, periods of the first input signal and the driving signal being both a predetermined period; each second drive circuit corresponding to a column of voltage integrators respectively, the each second drive circuit being configured to generate a column select signal according to a second input signal, the column select signal being used for controlling an operating state of a corresponding column of voltage integrators; and each of N columns of voltage integrators configured to receive the driving signal and the column select signal generated by a corresponding second drive circuit, and charge or integrate a fingerprint capacitor according to the driving signal and the column select signal.

Abstract: A fingerprint identification system includes: a fingerprint sensing circuit having a power supply terminal and a floating ground terminal, used for generating a first signal; and a signal generation circuit having a first output terminal and a second output terminal, wherein the first output terminal is coupled to the power supply terminal, and the second output terminal is coupled to the floating ground terminal, the signal generation circuit is used for generating a floating power signal to the power supply terminal according to the first signal, and for generating a floating ground signal to the floating ground terminal according to the first signal, wherein the floating power signal has a floating power amplitude, and the floating ground signal has a floating ground amplitude; and wherein the signal generation circuit has a breakdown voltage, and both the floating power amplitude and the floating ground amplitude are greater than the breakdown voltage.

Abstract: Provided are an under-screen biometric identification apparatus and an electronic device. The under-screen biometric identification apparatus includes: a lens disposed under a display screen for receiving an optical signal formed by reflection of a human finger on the display screen, where the optical signal is used to detect biometric information of the finger; a lens barrel, where the lens is fixed in the lens barrel; and a support, where the support is connected to the lens barrel by means of threaded connection for supporting the lens barrel. An under-screen biometric identification apparatus and an electronic device provided in embodiments of the present application can improve the efficiency of under-screen biometric identification.

Abstract: A device can include a fingerprint reader that generates fingerprint data representative of a fingerprint of a finger; circuitry that authenticates the fingerprint via an authentication process and that identifies a change of the fingerprint based at least in part on the generated fingerprint data and historical fingerprint data; and circuitry that, responsive to the identification of the change of the fingerprint, revises the authentication process based at least in part on at least a portion of the received fingerprint data.

Abstract: According to various example embodiments, an electronic device is disclosed. The electronic device includes a housing having a first surface facing a first direction and a second surface facing a second, opposing direction. A first area of the first surface includes a plurality of selectable input keys. A second area of the first surface excludes the plurality of keys. A sensor module, such as a fingerprint sensor, is installed to the first area.

Abstract: A security device includes a door handle unit disposed on a door body, a cover layer including a touch sensor and covering the door handle unit, a door locking unit which controls a locking of the door body. The touch sensor may include a plurality of sensor pixels, each including a sensor electrode which forms a second capacitor with a hand of a user when a touch occurs thereon by the user. The touch sensor may generate touch information based on a change in capacitance of the second capacitor corresponding to the touch.

Abstract: The subject disclosure presents systems and methods for receiving a plurality of assay information along with a query for one or more features of interest, and projecting anatomical information from an anatomical assay onto a staining assay, for example an immunohistochemical (IHC) assay that is commonly registered with the anatomical assay, to locate or determine features appropriate for analysis. The anatomical information may be used to generate a mask that is projected on one or more commonly registered staining assays. A location of the feature of interest in the staining assay may be correlated with the anatomical context provided by the mask, with any features of interest that match the anatomical mask being selected or indicated as appropriate for analysis.

Abstract: A computer-implemented method for recognizing a plurality of real-world toy construction elements of a toy construction system assembled to form a real-world toy construction model, each real-world toy construction element comprising coupling members for detachably connecting the real-world toy construction element with one or more other real-world toy construction elements of the toy construction system so as to form the real-world toy construction model; wherein the method comprises: receiving at least one captured image of the real world toy construction model; processing the at least one captured image so as to at least preliminarily recognise at least a first real-world toy construction element in the at least one captured image as a first known real-world toy construction element from stored digital representations of a plurality of known real-world toy construction elements of the toy construction system, each digital representation comprising connectivity information indicative of how the correspondin

Abstract: Described are techniques for indoor mapping and navigation. A reference mobile device including sensors to capture range, depth and position data and processes such data. The reference mobile device further includes a processor that is configured to process the captured data to generate a 2D or 3D mapping of localization information of the device that is rendered on a display unit, execute an object recognition to identify types of installed devices of interest of interest in a part of the 2D or 3D device mapping, integrate the 3D device mapping in the built environment to objects in the environment through capturing point cloud data along with 2D image or video frame data of the build environment.

Abstract: An eyebrow shaping method includes providing an image of a face. Multiple guidelines are determined on the image and points and areas are located via the guidelines. A face is shaped with contouring and lightening to have Golden Ratio proportions via the different points and zones.

Abstract: An image processing apparatus includes an imaging unit, a determination unit, and a discrimination unit. The imaging unit images a target object using signals in a predetermined wavelength band in which a spectral transmittance of a specific substance of a plurality of substances on a surface of the target object is different from a spectral transmittance of the other substances of the plurality of substances. The determination unit determines a distribution state of the specific substance on the surface of the target object based on an image captured by the imaging unit. The discrimination unit discriminates a class to which the target object belongs based on the distribution state of the specific substance.

Abstract: Systems and methods for identifying a false representation of a human face are provided. In one example, a method for identifying a false representation of a human face includes receiving one or more data streams captured by one or more sensors sensing a candidate face. In a plurality of stages that each comprises a different analysis, one or more of the data streams are analyzed, and the stages comprise determining whether a plurality of candidate face depth points lies on a single flat plane or a curving plane. Based at least in part on determining that the plurality of candidate face depth points lies on the single flat plane, an indication of the false representation of the human face is outputted.

Abstract: The present disclosure provides systems and methods that perform face reconstruction based on an image of a face. In particular, one example system of the present disclosure combines a machine-learned image recognition model with a face modeler that uses a morphable model of a human's facial appearance. The image recognition model can be a deep learning model that generates an embedding in response to receipt of an image (e.g., an uncontrolled image of a face). The example system can further include a small, lightweight, translation model structurally positioned between the image recognition model and the face modeler. The translation model can be a machine-learned model that is trained to receive the embedding generated by the image recognition model and, in response, output a plurality of facial modeling parameter values usable by the face modeler to generate a model of the face.

Abstract: A device for detecting and recording animal behavior is provided. The device includes at least one corral that defines a contained field, the base surface of the at least one corral being sensitive to the animal's footprint. The device also includes an image capturing device that cooperates with the base surface to capture a first video frame in which at least one of a tail, body and head of the animal is illuminated by a light below the animal and a second video frame in which both a profile of a full footprint and a profile of a toe print of the animal when the animal is standing on its toes is illuminated by a light that is totally internally reflected within the base surface. A control system is arranged to control the image capturing device to capture the first and second video frames.

Abstract: Devices, methods, and systems for occupancy detection are described herein. One device includes instructions to receive an image of a portion of a facility captured by an imaging device, the image defined by a field of view, and set a first occupancy detection threshold for a first part of the field of view and a second occupancy detection threshold for a second part of the field of view.

Abstract: Embodiments of the present disclosure pertain to extracting data from images using neural networks. In one embodiment, an image is fit to a predetermined bounding window. The image is then processed with a convolutional neural network to produce a three dimensional data cube. Slices of the cube are processed by an encoder RNN, and the results concatenated. The concatenated results are processed by an attention layer with input from a downstream decoder RNN. The attention layer output is provided to the decoder RNN to generate a probability array where values in the probability array correspond to particular characters in a character set. The maximum value is selected, and translated into an output character. In one embodiment, an amount may be extracted from an image of a receipt.

Abstract: The disclosure discloses a method, device and server of recognizing characters of a claim document, and a storage medium. The method includes: after the server receives a claim document image to be subjected to character recognition, performing region segmentation according to the frame line layout of a claim document frame format, thus obtaining one or multiple segmented regions; and calling a predetermined analysis model to analyze all the obtained segmented regions, and respectively performing character recognition on all the analyzed segmented regions by using a predetermined recognition rule to recognize characters in all the segmented regions.

Abstract: Approaches to produce verification are addressed which may be suitably employed in conjunction with a low resolution image scanner. Features, such as lines, corners, text, texture, edges and the like are extracted and compared with features for an identified item or evaluated to determine if the features collectively are indicative of a produce item or a non-produce item.

Abstract: In one example implementation according to embodiments of the invention, a computer-implemented method includes identifying a first discrete element of the composite object from a first image. The method further includes identifying the composite object based at least in part on the first discrete element of the composite object. The method further includes determining whether a category associated with the composite object is a category of interest. The method further includes, based at least in part on determining that the category associated with the composite object is a category of interest, increasing a discrete element count. The method further includes determining whether the discrete element count meets a threshold. The method further includes, based at least in part on determining that the discrete element count exceeds the threshold, performing a mitigation action.

Abstract: The application discloses an eyeball movement capturing method, which includes: acquiring a real-time image shot by a photographic device and extracting a real-time facial image; inputting the real-time facial image into a pretrained eyeball mean shape and recognizing n1 orbit feature points and n2 eyeball feature points; and calculating a movement direction and movement distance of an eyeball in the real-time facial image according to x and y coordinates of the (n1+n2) eyeball feature points in the real-time facial image. The application also discloses an electronic device and a computer-readable storage medium. According to the application, movement information of the eyeball in the real-time facial image is calculated according to the coordinates of the eye feature points to implement real-time capturing of an eyeball movement.

Abstract: A method for supporting visual tracking includes receiving a plurality of image frames captured at different times using an imaging device. Each of the plurality of image frames includes a plurality of pixels associated with a plurality of feature points. The method further includes analyzing the plurality of image frames to compute movement characteristics of the plurality of feature points and identifying a tracking feature relative to a background feature based on the movement characteristics of the plurality of feature points.

Abstract: A road detection method and apparatus. A specific embodiment of the method includes: acquiring an image of a predetermined region; semantically segmenting the image to acquire a first probability that a region corresponding to each pixel in the image is a road region; acquiring a historical position information set of a target terminal; correcting, in response to historical position information existing in the historical position information set, the historical position information indicating a historical position located in the predetermined region, the first probability according to the historical position information to obtain a second probability; and determining a region corresponding to a pixel having the second probability greater than a preset threshold as a road region. Such an embodiment improves the road detection accuracy.

Abstract: A computer implemented recognition process of an object in a query image provides a set of training images, each training image being defined by a plurality of pixels and comprising an object tag; determines for each training image of the set a plurality of first descriptors, each first descriptor being a vector that represents pixel properties in a corresponding subregion of the associated training image; and selects among the first descriptors a group of exemplar descriptors describing the set of training images, wherein selecting the exemplar descriptors includes determining the first descriptors having a number of repetitions in the set of training images higher than a certain value.