Sharp Kabushiki Kaisha Corporation

Real World Computing Partnership

Computer Basic Technology Research Association

Fumio Togawa

Toru Ueda

Yasushi Ishizuka

Hitoshi Hirose

Hitoshi Suzuki

Mitsuhiro Hakaridani

Takashi Aramaki

Yoichi Shimazawa

Kenji Sakamoto

Shin Kamiya

Hiroyuki Iwahashi

Atsuo Tanaka

Keiko Watanuki

A speech and moving image search apparatus includes an input section for inputting a command and a label information storage device for storing label information. A label attribute storage device is further included for storing label attributes, as well as a broader term storage device for storing broader terms of labels. A first search section extracts information on the data from the label information storage device. A database is also included for storing speech and moving images. A control section thereafter accesses desired data in the database based on the information extracted by the first search section. Finally, an output section is included for outputting the data.

Apparatus for searching for speech and moving images

When performing learning for a neural network, a plurality of learning vectors which belong to an arbitrary category are used, and self-organization learning in the category is carried out. As a result, the plurality of learning vectors which belong to the category are automatically clustered, and the contents of weight vectors in the neural network are set to representative vectors which exhibit common features of the learning vectors of each cluster. Then, teacher-supervised learning is carried out for the neural network, using the thus set contents of the weight vectors as initial values thereof. In the learning process, an initial value of each weight vector is set to the representative vector of each cluster obtained by clustering. Therefore, the number of calculations required until the teacher-supervised learning is converged is greatly reduced.

Neural network apparatus

A learning apparatus for use in a neural network system which has a plurality of classes representing different meanings. The learning apparatus is provided for learning a number of different patterns, inputted by input vectors, and classified in different classes. The learning apparatus is constructed by a computer and it includes a section for producing a plurality of output vectors representing different classes in response to an input vector, a section for obtaining a first largest output vector of all the output vectors, a section for obtaining a second largest output vector of all the output vectors, and a section for setting predetermined weights to the first and second largest output vectors, respectively, such that the first largest output vector is made larger, and the second largest output vector is made smaller. Furthermore, a section for determining a ratio of the weighted first and second largest output vectors, respectively, is included. If the determined ratio is smaller than a predetermined value, the weighted first and second largest output vectors are further weighted to be made further larger and smaller, respectively.

Neural network learning apparatus and method

A character recongintion device has a subdivider, a features calculator and a character code recognition device. Image data for a single character area is extracted from scanned character image data and input to the subdivider. This subdivider divides the image data for the single character area into subregions. The features calculator calculates quantified features in each subregion based on a degree of resemblance between a template and image data in the subregions. When the features of each subregion are calculated for all subregions constituting the single character area, a character code corresponding to the scanned character image data is recognized by the character code recognition device based on the quantified features of each of all subregions.

Character recognition device which divides a single character region

A neural network system and method of operating same wherein input data are initialized, then mapped onto a predetermined array for learning or recognition. The mapped information is divided into sub-input data or receptive fields, which are used for comparison of the input information with prelearned information having similar features, thereby allowing for correct classification of the input information. The receptive fields are shifted before the classification process, in order to generate a closest match between features which may be shifted at the time of input, and weights of the input information are updated based upon the closest-matching shifted position of the input information.

Receptive field neural network with shift-invariant pattern recognition

A control device for controlling the learning of a neural netowrk includes a monitor for monitoring weight values of synapse connections between units of the neural netowrk during learning of the neural network so as to update these weight values. When one of the weight values satisfies a preset condition, the weight value is updated to a predetermined value such that configuration of the neural network is determined in an optimum manner.

Control device for controlling learning of a neural network

A character recognition device has a subdivider, a features calculator and a character code recognition device. Image data for a single character area is extracted from scanned character image data and input to the subdivider. This subdivider divides the image data for the single character area into subregions. The features calculator calculates quantified features in each subregion based on a degree of resemblance between a template and image data in the subregions. When the features of each subregion are calculated for all subregions constituting the single character area, a character code corresponding to the scanned character image data is recognized by the character code recognition device based on the quantified features of each of all subregions.

A figure input system includes an input device such as a digitizer through which coordinate data of points on a figure to be inputted are time-wise continuously received. These data are sampled sequentially at set time intervals and are analyzed in view of the kind of basic figure intended by the user such that geometrical characteristics and position data are extracted from the sampled data and a regular basic figure can be displayed on a display device.

Figure input system

A recognizing apparatus is provided for recognizing a class to which an inputted characteristic pattern belongs from among a plurality of classes to be discriminated using a neural network. The classes are classified into a plurality of categories. The apparatus includes a network selecting portion for selecting a category to which the inputted characteristic pattern belongs and for selecting a neural network for use in discriminating the class to which the inputted characteristic pattern belongs in the selected category. The apparatus further includes a network memory portion, a network setting portion and a details discriminating portion. The network memory portion stores structures of a plurality of neural networks which have finished learning for respective categories, weights of the neural networks set by the learning and a plurality of discriminating algorithms to be used when the classes are discriminated by the neural networks. The network setting portion sets the structure and weights of a neural network selected by the network selecting portion and a discriminating alogrithm appropriate to the selected category. The details discriminating portion recognizes the class to which the inputted characteristic pattern belongs by performing the details discriminating operation using the neural network set by the neural network setting portion.

Recognizing apparatus

A character recognition apparatus includes an inputting function of ruled lines. The apparatus is equipped with a tablet type input device integrating a tablet type input device with a display panel. The apparatus includes a control unit that recognizes information inputted in an input panel by a stylus pen. There is a memory unit for storing the recognized information. The control unit sets a grating-like virtual frame in the input panel. The frame set in the input panel is a virtual boarder obtained by calculation and does not appear on a display panel.

Character recognition apparatus with inputting function of ruled lines

A self-extending shape neural-network is capable of a self-extending operation in accordance with the studying results. The self-extending shape neural-network has initially minimum number of the intermediate layers and the number of the nodes (units) within each layer by the self-extension of the network construction so as to shorten the studying time and the discriminating time. This studying may be effected efficiently by the studying being directed towards the focus when the studying is not focused.

Self-extending neural-network

An information recognition system includes a recognition unit for recognizing inputted information as groups of candidate characters, a memory for storing the groups of candidate characters, a display-integrated tablet for displaying top characters in the groups of candidate characters, also a coordinate input device for indicating one of the displayed characters, and a character correcting device which replaces the top character indicated by the coordinate indicating device with another character in the group to which the indicated top character belongs are provided.

Information recognition system

A handwritten character-recognizing apparatus having a discriminating device which discriminates whether or not an input device, namely, the pen of a writing/displaying tablet has moved from one character frame to the following character frame. When it is discriminated that the input device has moved to the following character frame, a character frame generating device automatically generates a character frame adjacent to the last character frame of the character frames arranged in a line. Thereby, the reading of handwritten characters and the recognizing of standard characters which are displayed instead of the handwritten characters are facilitated.

Handwritten character-recognizing apparatus for automatically generating

Inputted voice signals are analyzed in units of syllables. Each syllable is compared with standard syllables preregistered in a memory and thereby recognized but corrections may be made on erroneous recognitions by referencing a dictionary or entering a command. Each standard pattern is associated with phonological information on the neighborhood in a voice signal from where it was extracted and may be updated by another pattern having the same phonological information. Temporal sequences of correct and erroneous recognitions of individual syllables as well as whole syllables are stored and referenced in determining whether a standard pattern should be updated. A maximum pattern number may be set for each syllable.

Voice recognition method and apparatus by updating reference patterns

Spoken words are recognized in a sequence of steps: first finding candidate words in the form of mono-syllabic sequences; then associating (correcting) each mono-syllabic sequence to more speech-like sounds, then comparing with pre-stored dictionary words for final recognition.

Growing community of inventors

Nara, Japan

Fumio Togawa