摘要:
Processing data which belong to different classes and label data indicating the classes to which the processing data belong are input (S20). A distance relationship between the processing data is calculated (S22). An interclass separation degree between the classes is set (S23). The distance relationship is updated based on the label data and the interclass separation degree (S24). A data mapping relationship which approximates the updated distance relationship is calculated (S25).
摘要:
A plurality of pieces of learning data, each associated with a class to which the piece of the learning data belong, are input. In each piece of the learning data, a statistical amount of attribute values of elements in each of specific k parts, k being equal to or larger than 1, is calculated. Each piece of the learning data is mapped in a k-dimensional feature space as a vector having the calculated k statistics amounts as elements. Based on each piece of the mapped learning data and the classes to which the pieces of learning data belong, parameters for classifying input data into one of the plurality of classes are learned in the k-dimensional feature space. By using the parameters, pattern classification can be performed with high speed and high accuracy.
摘要:
A hierarchical processing apparatus includes a data input unit for time-sequentially inputting pattern data in a particular hierarchical level, a calculation parameter control data memory for storing calculation parameter control data, a local area recognition module for detecting a particular feature from pattern data, using the calculation parameter control data, and an intermediate result storage module for storing an intermediate result output from the local area recognition module. Depending on the hierarchical level of processing, a flow of data is controlled such that an input acquired via the data input unit or an intermediate result fed back from the intermediate result storage module and calculation parameter control data read from the calculation parameter control data memory are supplied to the local area recognition module. The hierarchical processing apparatus can perform various kinds of complicated calculations using a simple small-scale circuit with low power consumption.
摘要:
In a hierarchical neural network having a module structure, learning necessary for detection of a new feature class is executed by a processing module which has not finished learning yet and includes a plurality of neurons which should learn an unlearned feature class and have an undetermined receptor field structure by presenting a predetermined pattern to a data input layer. Thus, a feature class necessary for subject recognition can be learned automatically and efficiently.
摘要:
An image processing apparatus stores model information representing a subject model belonging to a specific category, detects the subject from an input image by referring to the model information, determines a region for which an image correction is to be performed within a region occupied by the detected subject in the input image, stores, for a local region of the image, a plurality of correction data sets representing correspondence between a feature vector representing a feature before correction and a feature vector representing a feature after correction, selects at least one of the correction data sets to be used to correct a local region included in the region determined to undergo the image correction, and corrects the region determined to undergo the image correction using the selected correction data sets.
摘要:
An apparatus includes a capturing unit that captures a target object and generates a range image representing distance information, a general estimation unit that analyzes the range image and estimates a general position and orientation of the target object, and a determination unit that determines a priority order of a plurality of identification units, based on the estimated general position and orientation. In addition, a setting unit sets a search window indicating a range for detailed estimation of position and orientation of the range image based on a preregistered general size of the target object, a calculation unit calculates a difference between a first distance value of a first pixel in the range image and a second distance value of a second pixel which is next to the first pixel in the range image, and an updating unit updates the search window based on the calculated difference. A detailed estimation unit estimates a detailed position and orientation of the target object, using the plurality of identification units in the determined priority order within the range of the search window.
摘要:
A feature selection apparatus, which selects features to be used to discriminate an object by a discriminator using learning data including the object, extracts a plurality of partial data from the learning data, and obtains discrimination values obtained by controlling the discriminator to process the plurality of extracted partial data as features of the plurality of partial data. The feature selection apparatus evaluates the obtained features based on discrimination degrees on a discrimination space defined by the discriminator, and selects features to be used to discriminate the object from a plurality of features obtained in association with the plurality of partial data based on an evaluation result.
摘要:
A pattern identification apparatus for identifying one of a plurality of classes defined in advance, to which data of a pattern identification target belongs, comprises a read unit adapted to read out, from a storage unit in correspondence with each of the plurality of classes, a projection rule to a hyperplane which approximates a manifold corresponding to the class in a feature space an input unit adapted to input identification target data; a calculation unit adapted to calculate, for each class, a projection result obtained by projecting the input identification target data to the hyperplane which approximates the manifold corresponding to each of the plurality of classes, on the basis of the projection rule; and an identification unit adapted to identify, on the basis of the projection result of each classes calculated by said calculation unit, one of the plurality of classes to which the identification target data belongs.
摘要:
Processing data which belong to different classes and label data indicating the classes to which the processing data belong are input (S20). A distance relationship between the processing data is calculated (S22). An interclass separation degree between the classes is set (S23). The distance relationship is updated based on the label data and the interclass separation degree (S24). A data mapping relationship which approximates the updated distance relationship is calculated (S25).
摘要:
A plurality of pieces of learning data, each associated with a class to which the piece of the learning data belong, are input. In each piece of the learning data, a statistical amount of attribute values of elements in each of specific k parts, k being equal to or larger than 1, is calculated. Each piece of the learning data is mapped in a k-dimensional feature space as a vector having the calculated k statistics amounts as elements. Based on each piece of the mapped learning data and the classes to which the pieces of learning data belong, parameters for classifying input data into one of the plurality of classes are learned in the k-dimensional feature space. By using the parameters, pattern classification can be performed with high speed and high accuracy.