Abstract: A method of tracking an object across a stream of images comprises determining a region of interest (ROI) bounding the object in an initial frame of an image stream. A HOG map is provided for the ROI by: dividing the ROI into an array of M×N cells, each cell comprising a plurality of image pixels; and determining a HOG for each of the cells. The HOG map is stored as indicative of the features of the object. Subsequent frames are acquired from the stream of images. The frames are scanned ROI by ROI to identify a candidate ROI having a HOG map best matching the stored HOG map features. If the match meets a threshold, the stored HOG map indicative of the features of the object is updated according to the HOG map for the best matching candidate ROI.

Abstract: Aspects of the present disclosure are directed to techniques that improve performance of CNN systems through the effect of improved memory efficiencies for CNNs operating on GPUs. Aspects of the disclosure demonstrate that off-chip memory in such CNN systems is underutilized due to at least three characteristics namely, data layout, data locality and inter-kernel redundancy. Aspects of the disclosure examine the performance impact of different data layouts and then describe a method to produce data layout selection for various layers of the CNN including a fast transformation implementation. Disclosed are improvements to data locality from working set expansion, elimination of inter-kernel redundancy and increase of TLP using kernel reconstruction techniques including kernel fusion and thread injection. Disclosed experimental results show that our optimizations are very effective to boost the performance of CNNs by amounts up to 9.76 times for a single kernel and 2.05 times for a network.

Abstract: In an example, a first machine learning algorithm is used to train a smart contour model to identify contours of product shapes in input images and to identify backgrounds in the input images. A second machine learning algorithm is used to train a plurality of shape-specific classification models to output identifications of products in input images. A candidate image of one or more products is obtained. The candidate image is passed to the smart contour model, obtaining output of one or more tags identifying product contours in the candidate image. The candidate image and the one or more tags are passed to an ultra-large scale multi-hierarchy classification system to identify one or more classification models for one or more individual product shapes in the candidate image. The one or more classification models are used to distinguish between one or more products and one or more unknown products in the image.

Abstract: A system and method for invoice field detection and parsing includes the steps of extracting character bounding blocks using optical character recognition (OCR) or digital character extraction (DCE), enhancing the image quality, analyzing the document layout based on imaging techniques, detecting the invoice field based on the machine learning techniques, and parsing the invoice field value based on the content information.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating descriptions of input images. One of the methods includes obtaining an input image; processing the input image using a first neural network to generate an alternative representation for the input image; and processing the alternative representation for the input image using a second neural network to generate a sequence of a plurality of words in a target natural language that describes the input image.

Abstract: A system and method of generating a natural language generation (NLG) output, wherein the method includes: receiving speech signals from a user at a microphone of a client device; determining a requested communication goal and at least one inputted communication value based on the received speech signals; determining to use a static natural language generation (NLG) template or a dynamic NLG template to generate an NLG output, wherein the determination of whether to use a static NLG template or a dynamic NLG template is made using a neural network NLG template selection process; selecting an NLG template after the determination of whether to use a static NLG template or a dynamic NLG template; and generating an NLG output based on the selected NLG template.

Abstract: System and method for detecting objects in geospatial images, 3D point clouds and Digital Surface Models (DSMs). Deep Convolution Neural Networks (DCNNs) are trained using positive and negative training examples. Using a rotation pattern match of only positive examples reduces the number of negative examples required. In DCNNs softmax probability is variant of rotation angles. When rotation angle is coincident with object orientation, softmax probability has maximum value. During training, positive examples are rotated so that their orientation angles are zero. During detection, test images are rotated through different angles. At each angle, softmax probability is computed. A final object detection is based on maximum softmax probability as well as a pattern match between softmax probability patterns of all positive examples and the softmax probability pattern of a target object at different rotation angles.

Abstract: Operations of a combination of first and second original convolutional layers followed by a short path are replaced by operations of a set of three particular convolutional layers. The first contains 2N×N filter kernels formed by placing said N×N filter kernels of the first original convolutional layer in left side and N×N filter kernels of an identity-value convolutional layer in right side. The second contains 2N×2N filter kernels formed by placing the N×N filter kernels of the second original convolutional layer in upper left corner, N×N filter kernels of an identity-value convolutional layer in lower right corner, and N×N filter kernels of two zero-value convolutional layers in either off-diagonal corner. The third contains N×2N of kernels formed by placing N×N filter kernels of a first identity-value convolutional layer and N×N filter kernels of a second identity-value convolutional layer in a vertical stack. Each filter kernel contains 3×3 filter coefficients.

Abstract: A method of decoding a received message includes: determining a weighting vector corresponding to at least one bit of the received message according to a syndrome and a parity check matrix; determining a bit state of the bit according to a bit value of the bit; changing the bit state according to the weighting vector and a flipping threshold, wherein a change range of the bit state is variable; and flipping the bit according to the bit state.

Abstract: A method implemented by a processor for coding a real signal, for example an image signal, into a quantized signal, comprises the following steps applied to each real sample of the real signal: converting the real sample into a digital representation, selecting, in the fractional part of the number, a predetermined number N of most significant non-zero bits, for each non-zero significant bit i selected, i varying from 1 to N, determining its distance Pi with respect to the neighboring selected non-zero significant bit of higher rank or, for the first non-zero significant bit selected, with respect to the decimal point, deducting from the distance Pi the minimum value of distance between two non-zero bits, coding the modified distance Pi on a predetermined number Mi of bits.

Abstract: Medicine packaging apparatuses and methods for accurately determining a remaining sheet amount of a medicine packaging sheet are described. The apparatus includes: a roll support section to which a core tube of a medicine packaging sheet roll is attached; a sensor disposed in the roll support section for outputting a count value according to a rotation amount; a wireless reader-writer unit for writing information to a core tube IC tag and reading said information; an information generation section for generating information to be written to the core tube IC tag; a remaining sheet amount estimation section for estimating a current amount of remaining sheet based on the information and dimensional information of the core tube; and a controller which selectively performs an operation if a reference time-point count value is not yet written to the core tube IC tag and another operation if the count value is already written thereto.

Abstract: A string of Latin-alphabet based language texts is received and formed a multi-layer 2-D symbol in a computing system. The received string contains at least one word with each word containing at least one letter of the Latin-alphabet based language. 2-D symbol comprises a matrix of N×N pixels of data representing a super-character. The matrix is divided into M×M sub-matrices. Each sub-matrix represents one ideogram formed from the at least one letter contained in a corresponding word in the received string. Ideogram has a square format with a dimension EL letters by EL letters (i.e., row and column). EL is determined from the total number of letters (LL) contained in the corresponding word. EL, LL, N and M are positive integers. Super-character represents a meaning formed from a specific combination of at least one ideogram. Meaning of the super-character is learned with image classification of the 2-D symbol.

Abstract: A human-shape image segmentation method comprising: extracting multi-scale context information for all first pixel points for training a human-shape image; sending image blocks of all scales of all the first pixel points into a same convolution neural network to form a multi-channel convolutional neural network group, wherein each channel corresponds to image blocks of one scale; training the neural network group using a back propagation algorithm to obtain human-shape image segmentation training model data; extracting multi-scale context information for all second pixels points for testing the human-shape image; sending image blocks of different scales of each of the second pixel points into a neural network channel corresponding to the human-shape image segmentation training model, wherein if said first probability is larger than said second probability, the second pixel points belong to the human-shape region, otherwise, the second pixel points are outside of the human-shape region.

Abstract: A method is described including storing reference vector data corresponding to user gestures at a plurality of neurons at pattern matching hardware, receiving real time signals from the sensor array and performing gesture recognition using the pattern matching hardware to compare incoming vector data corresponding to the real time signals with the reference vector data.

Abstract: A method for image processing includes determining features of multiple stored images from a pre-trained deep convolutional network. The method also includes clustering each image of the multiple stored images based on the determined features.

Abstract: Imaging processing techniques using a trained convolution neural network (CNN) are described. In one or more implementations, an image processing system and method are provided for applying an image processing algorithm to a dataset of training images to generate a plurality of performance curves, constructing a loss function based upon the plurality of performance curves, training a convolutional neural network (CNN) to optimize the loss function to establish a trained convolutional neural network, predicting a specific tuning parameter for an image of interest using the trained convolution neural network and performing image processing of the image of interest using the specific tuning parameter and the image processing algorithm to generate a processed image of interest.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying objects in images. One of the methods includes obtaining data identifying a plurality of object categories; obtaining relationship data for the plurality of object categories; determining a plurality of valid assignments from the relationship data; obtaining, for an image, a respective raw image score for each of the object categories, wherein the raw image score for each of the object categories represents an estimate of a likelihood that the image includes an image of an object that belongs to the object category; and generating a respective final score for each of the object categories from the raw image scores using the valid assignments, wherein the final score for each of the object categories represents the likelihood that the image includes an image of an object that belongs to the object category.

Abstract: A Convolutional Neural Network (CNN) includes an initial set of convolutional layers and max pooling units, in which any input is convoluted with the learned image filters and the output is a stack of the different filter responses. Max pooling produces a scaled version of the output. The process can be repeated several times, resulting in a stack of space invariant-scaled images. Since the operation is space invariant, the computations of these layers not need to be recomputed if interested just in certain regions of the image. A Region Of Interest (ROI) Pooling layer is used to select regions to be processed by the set of fully connected layers, which uses the response of the multiple convolutional layers of the network to determine the regions where the objects (of different scales) could be located. This object proposal method is implemented as a Region Of Interest (ROI) Selector.

Abstract: The invention concerns a method for the computerized control and/or regulation of a technical system. Within the context of the method according to the invention, an action-selection rule (PO?) is determined which has a low level of complexity and yet is well suited to the regulating and/or control of the technical system, there being used for determination of the action-selection rule (PO?) an evaluation measure (EM) which is determined on the basis of a distance measure and/or a reward measure and/or an action-selection rule evaluation method. The action-selection rule is then used to control and/or regulate the technical system. The method according to the invention has the advantage of the action-selection rule being comprehensible to a human expert. Preferably, the method according to the invention is used for regulating and/or controlling a gas turbine and/or a wind turbine.

Abstract: Embodiments of the invention relate to canonical spiking neurons for spatiotemporal associative memory. An aspect of the invention provides a spatiotemporal associative memory including a plurality of electronic neurons having a layered neural net relationship with directional synaptic connectivity. The plurality of electronic neurons configured to detect the presence of a spatiotemporal pattern in a real-time data stream, and extract the spatiotemporal pattern. The plurality of electronic neurons are further configured to, based on learning rules, store the spatiotemporal pattern in the plurality of electronic neurons, and upon being presented with a version of the spatiotemporal pattern, retrieve the stored spatiotemporal pattern.

Abstract: Systems, apparatuses, and methods for k-nearest neighbor (KNN) searches are described. In particular, embodiments of a KNN accelerator and its uses are described. In some embodiments, the KNN accelerator includes a plurality of vector partial distance computation circuits each to calculate a partial sum, a minimum sort network to sort partial sums from the plurality of vector partial distance computation circuits to find k nearest neighbor matches and a global control circuit to control aspects of operations of the plurality of vector partial distance computation circuits.

Abstract: Embodiments of the invention relate to canonical spiking neurons for spatiotemporal associative memory. An aspect of the invention provides a spatiotemporal associative memory including a plurality of electronic neurons having a layered neural net relationship with directional synaptic connectivity. The plurality of electronic neurons configured to detect the presence of a spatiotemporal pattern in a real-time data stream, and extract the spatiotemporal pattern. The plurality of electronic neurons are further configured to, based on learning rules, store the spatiotemporal pattern in the plurality of electronic neurons, and upon being presented with a version of the spatiotemporal pattern, retrieve the stored spatiotemporal pattern.

Abstract: An approach to detecting objects in an image dataset may combine texture/color detection, shape/contour detection, and/or motion detection using sparse, generative, hierarchical models with lateral and top-down connections. A first independent representation of objects in an image dataset may be produced using a color/texture detection algorithm. A second independent representation of objects in the image dataset may be produced using a shape/contour detection algorithm. A third independent representation of objects in the image dataset may be produced using a motion detection algorithm. The first, second, and third independent representations may then be combined into a single coherent output using a combinatorial algorithm.

Abstract: An exemplary methodology, procedure, system, method and computer-accessible medium can be provided for receiving physiological data for the subject, extracting one or more patterns of features from the physiological data, and classifying the at least one state of the subject using a spatial structure and a temporal structure of the one or more patterns of features, wherein at least one of the at least one state is an ictal state.

Abstract: Provided is a characteristic point associating system including: a set creating unit to receive a plurality of characteristic point groups to be compared, and to create a plurality of characteristic point pair sets by grouping together characteristic point pairs that are close to one another in terms of local conversion parameter into sets; a set selecting unit to select a characteristic point pair set that contains many elements out of the plurality of characteristic point pair sets; and a corresponding characteristic point determining unit to determine out of characteristic point pairs contained in the selected characteristic point pair set, a pair of characteristic points to be associated with each other as correct corresponding characteristic points so as to be output. Thus, the characteristic point associating system associates correct pairing combinations of characteristic points that exist between the compared groups of characteristic points.

Abstract: Embodiments of the invention relate to a scalable neural hardware for the noisy-OR model of Bayesian networks. One embodiment comprises a neural core circuit including a pseudo-random number generator for generating random numbers. The neural core circuit further comprises a plurality of incoming electronic axons, a plurality of neural modules, and a plurality of electronic synapses interconnecting the axons to the neural modules. Each synapse interconnects an axon with a neural module. Each neural module receives incoming spikes from interconnected axons. Each neural module represents a noisy-OR gate. Each neural module spikes probabilistically based on at least one random number generated by the pseudo-random number generator unit.

Abstract: The present technique includes: an area-characteristic detector configured to calculate a maximum value, an average value, and a minimum value of signal levels of pixels around a certain pixel; a first gain creating part configured to calculate a first calculation value for the certain pixel; a second gain creating part configured to calculate a second calculation value for the certain pixel; and an adjustment part configured; to perform enhancement by multiplying the difference between the average value and the signal level of the certain pixel by the first calculation value when the signal level of the certain pixel is equal to or higher than the average value, and to perform enhancement by multiplying the difference between the average value and the signal level of the certain pixel by the second calculation value when the signal level of the certain pixel is lower than the average value.

Abstract: In an embodiment, a recognition apparatus includes an obtaining unit, a calculating unit, a principal axis selecting unit, a turning point setting unit, a section setting unit, and a determining unit. The obtaining unit obtains positions of a specific part in a coordinate system having a first axis to an n-th axis (n?2). The calculating unit calculates a movement vector of the specific part. The principal axis selecting unit selects a principal axis. The turning point setting unit sets a position at which there is a change in the principal axis and sets a position at which there is a change. The section setting unit sets a determination target section, and sets a previous section. The determining unit calculates an evaluation value of the determination target section and an evaluation value of the immediately previous section and determines which of the first axis to the n-th axis is advantageous.

Abstract: A learning apparatus in the present invention includes a weak discriminator generation unit that generates a weak discriminator which calculates a discrimination score of an instance of a target based on a feature and a bag label, a weak discrimination unit which calculates the discrimination score based on the generated weak discriminator, an instance probability calculation unit that calculates an instance probability of the target instance based on the calculated the discrimination score, a bag probability calculation unit that calculates a probability that no smaller than two positive instances are included in the bag based on the calculated instance probability and a likelihood calculation unit which calculates likelihood representing plausibility of the bag probability based on the bag label.

Abstract: Provided us a visual cortical circuit apparatus comprising: a current mirror unit which uses a transistor as a current source to generate a current having the same size as that of a reaction; a transconductance unit which takes, as an input, the current generated by the current mirror unit and outputs a voltage using a transconductance; and a buffer unit for converting the voltage output from the transconductance unit into a current and buffering the current.

Abstract: A method of testing a video against an aggregate query includes automatically receiving an aggregate query defining participant(s) and condition(s) on the participant(s). Candidate object(s) are detected in the frames of the video. A first lattice is constructed for each participant, the first-lattice nodes corresponding to the candidate object(s). A second lattice is constructed for each condition. An aggregate lattice is constructed using the respective first lattice(s) and the respective second lattice(s). Each aggregate-lattice node includes a scoring factor combining a first-lattice node factor and a second-lattice node factor. respective aggregate score(s) are determined of one or more path(s) through the aggregate lattice, each path including a respective plurality of the nodes in the aggregate lattice, to determine whether the video corresponds to the aggregate query.

Abstract: Apparatus and methods for contrast enhancement and feature identification. In one implementation, an image processing apparatus utilizes latency coding and a spiking neuron network to encode image brightness into spike latency. The spike latency is compared to a saliency window in order to detect early responding neurons. Salient features of the image are associated with the early responding neurons. A inhibitory neuron receives salient feature indication and provides inhibitory signal to the other neurons within an area of influence of the inhibitory neuron. The inhibition signal reduces probability of responses by the other neurons to stimulus that is proximate to the feature thereby increasing contrast within the encoded data. The contrast enhancement may facilitate feature identification within the image. Feature detection may be used for example for image compression, background removal and content distribution.

Abstract: The invention includes a method for recognizing shapes using a preprocessing mechanism that decomposes a source signal into basic components called atoms and a recognition mechanism that is based on the result of the decomposition performed by the preprocessing mechanism. In the method, the preprocessing mechanism includes at least one learning phase culminating in a set of signals called kernels, the kernels being adapted to minimize a cost function representing the capacity of the kernels to correctly reconstruct the signals from the database while guaranteeing a sparse decomposition of the source signal while using a database of signals representative of the source to be processed and a coding phase for decomposing the source signal into atoms, the atoms being generated by shifting of the kernels according to their index, each of the atoms being associated with a decomposition coefficient. The invention also includes a shape recognition system for implementing the method.

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: Systems and methods for automated pattern recognition and object detection. The method can be rapidly developed and improved using a minimal number of algorithms for the data content to fully discriminate details in the data, while reducing the need for human analysis. The system includes a data analysis system that recognizes patterns and detects objects in data without requiring adaptation of the system to a particular application, environment, or data content. The system evaluates the data in its native form independent of the form of presentation or the form of the post-processed data.

Abstract: An image-based biomarker is generated using image features obtained through object-oriented image analysis of medical images. The values of a first subset of image features are measured and weighted. The weighted values of the image features are summed to calculate the magnitude of a first image-based biomarker. The magnitude of the biomarker for each patient is correlated with a clinical endpoint, such as a survival time, that was observed for the patient whose medical images were analyzed. The correlation is displayed on a graphical user interface as a scatter plot. A second subset of image features is selected that belong to a second image-based biomarker such that the magnitudes of the second image-based biomarker for the patients better correlate with the clinical endpoints observed for those patients. The second biomarker can then be used to predict the clinical endpoint of other patients whose clinical endpoints have not yet been observed.

Abstract: Described herein is a technology for facilitating deformable model-based segmentation of image data. In one implementation, the technology includes receiving training image data (202) and automatically constructing a hierarchical structure (204) based on the training image data. At least one spatially adaptive boundary detector is learned based on a node of the hierarchical structure (206).

Abstract: In a method for determining a parameter in an automatic study and data management system, data is gathered in a knowledge database, and a parameter is determined based the data gathered in the knowledge database. The data is correlated to at least one of a configuration and implementation of a previous clinical study. The parameter is usable for configuring a future clinical study.

Abstract: A plurality of features determined from at least a portion of an image containing information about an object are processed with an inclusive neural network, and with a plurality of exclusive neural networks, so as to provide a plurality of inclusive probability values representing probabilities that the portion of the image corresponds to at least one of at least two different classes of objects, and for each exclusive neural network, so as to provide first and second exclusive probability values representing probabilities that the portion of the image respectively corresponds. or not. to at least one class of objects. The plurality of inclusive probability values, and the first and second exclusive probability values from each of the exclusive neural networks, provide for identifying whether the portion of the image corresponds, or not, to any of the at least two different classes of objects.

Abstract: Data set generation and data set presentation for image processing are described. The processing determines a location for each of one or more musical artifacts in the image and identifies a corresponding label for each of the musical artifacts, generating a training file that associates the identified labels and determined locations of the musical artifacts with the image, and presenting the training file to a neural network for training.

Abstract: Systems and methods for implementing a multi-label image recognition framework for classifying digital images are provided. The provided multi-label image recognition framework utilizes an iterative, multiple analysis path approach to model training and image classification tasks. A first iteration of the multi-label image recognition framework generates confidence maps for each label, which are shared by the multiple analysis paths to update the confidence maps in subsequent iterations. The provided multi-label image recognition framework permits model training and image classification tasks to be performed more accurately than conventional single-label image recognition frameworks.

Abstract: A method and system for training a neural network of a visual recognition computer system, extracts at least one feature of an image or video frame with a feature extractor; approximates the at least one feature of the image or video frame with an auxiliary output provided in the neural network; and measures a feature difference between the extracted at least one feature of the image or video frame and the approximated at least one feature of the image or video frame with an auxiliary error calculator. A joint learner of the method and system adjusts at least one parameter of the neural network to minimize the measured feature difference.

Abstract: Various technologies and techniques are disclosed that improve handwriting recognition operations. Handwritten input is received in training mode and run through several base recognizers to generate several alternate lists. The alternate lists are unioned together into a combined alternate list. If the correct result is in the combined list, each correct/incorrect alternate pair is used to generate training patterns. The weights associated with the alternate pairs are stored. At runtime, the combined alternate list is generated just as training time. The trained comparator-net can be used to compare any two alternates in the combined list. A template matching base recognizer is used with one or more neural network base recognizers to improve recognition operations. The system provides comparator-net and reorder-net processes trained on print and cursive data, and ones that have been trained on cursive-only data. The respective comparator-net and reorder-net processes are used accordingly.

Abstract: An intrusion prevention/detection system filter (IPS filter) performance evaluation is provided. The performance evaluation is performed at both the security center and at the customer sites to derive a base confidence score and local confidence scores. Existence of new vulnerability is disclosed and its attributes are used in the generation of new IPS filter or updates. The generated IPS filter is first tested to determine its base confidence score from test confidence attributes prior to deploying it to a customer site. A deep security manager and deep security agent, at the customer site, collect local confidence attributes that are used for determining the local confidence score. The local confidence score and the base confidence score are aggregated to form a global confidence score. The local and global confidence scores are then compared to deployment thresholds to determine whether the IPS filter should be deployed in prevention or detection mode or sent back to the security center for improvement.

Abstract: A disclosed image processing apparatus includes process components configured to input, process, or output image data; and a registration unit configured to obtain a list of the process components available in the image processing apparatus, to display on a display unit a screen for selecting a plurality of the process components from the list and thereby defining a combination of the process components which combination implements an application program for performing an image processing task, and to register the combination with an identifier in the image processing apparatus.

Abstract: A plurality of images inputted in an image signal input portion are divided into a plurality of regions by an image dividing portion, and a feature value in each of the plurality of regions is calculated by a feature value calculation portion and divided into a plurality of subsets by a subset generation portion. On the other hand, a cluster classifying portion classifies a plurality of clusters generated in a feature space into any one of a plurality of classes on the basis of the feature value and occurrence frequency of the feature value. And a classification criterion calculation portion calculates a criterion of classification for classifying images included in one subset on the basis of a distribution state of the feature value in the feature space of each of the images included in the one subset.

Abstract: An image-based pattern recognizer and a method and apparatus for making such a pattern recognizer are disclosed. By employing positional coding, the meaning of any feature present in an image can be defined implicitly in space. The pattern recognizer can be a neural network including a plurality of stages of observers. The observers are configured to cooperate to identify the presence of features in the input image and to recognize a pattern in the input image based on the features. Each of the observers includes a plurality of neurons. The input image includes a plurality of units, and each of the observers is configured to generate a separate output set that includes zero or more coordinates of such units.

Abstract: An image-based pattern recognizer and a method and apparatus for making such a pattern recognizer are disclosed. By employing positional coding, the meaning of any feature present in an image can be defined implicitly in space. The pattern recognizer can be a neural network including a plurality of stages of observers. The observers are configured to cooperate to identify the presence of features in the input image and to recognize a pattern in the input image based on the features. Each of the observers includes a plurality of neurons. The input image includes a plurality of units, and each of the observers is configured to generate a separate output set that includes zero or more coordinates of such units.

Abstract: A method of compression of videotelephony images characterized by: creating (10) a learning base containing images; centering the learning base about zero; determining component images by principal component analysis (12); and keeping a number of significant principal components (14).