Abstract: A method and system for automatically modifying an original transcription produced as the output of a recognition operation produces a second, modified transcription, such as, for example, automatically correcting an errorful transcription produced by an OCR operation. The invention uses information in an input text image of character images and in an original transcription associated with the input text image to modify aspects of a formal image source model that models as a grammar the spatial image structure of a set of text images. A recognition operation is then performed on the input text image using the modified formal image source model to produce a second, modified transcription. When the original transcription is errorful, the second transcription is a corrected transcription. Several aspects of the formal image source model may be modified; in particular, character templates to be used in the recognition operation are trained in the font of the glyphs occurring in the input text image.

Abstract: A method for segmenting audio data, comprising speech from a plurality of individual speakers, according to speaker is provided. The method comprises providing individual HMMs for each individual speaker, each individual HMM including at least one state, and constructing a speaker network HMM by connecting the individual HMMs in parallel. The audio data is then divided into segments by determining a most likely sequence of states through the speaker network HMM, each of the segments being associated with one of the individual HMMs. Afterward, the speaker of each of the segments is identified. The segmented data may be used to form an index into the audio data according to speaker.

Abstract: A processor controlled system for correlating an electronic index according to speaker for audio data being recorded in real time. The system includes a source of training data for each of the plurality of individual speakers and audio input system for providing real time audio data including speech for the individual speakers. The audio data is converted into spectral feature data by an audio processor, and is simultaneously recorded on a storage medium by a recording device. A system processor accepts the training data to create individual speaker models, which are combined in parallel to form a speaker network. The system processor then accepts the spectral feature data of the audio data and, using the speaker network, determines segments in the audio data corresponding to each speaker.

Abstract: A weighted wavelet hierarchical vector quantization (WWHVQ) procedure is initiated by obtaining an N.times.N pixel image where 8 bits per pixel (steps 10 and 12). A look-up operation is performed to obtain data representing a discrete wavelet transform (DWT) followed by a quantization of the data (step 14). Upon completion of the look-up, a data compression will have been performed. Further stages and look-up will result in further compression of the data, i.e., 4:1, 8:1, 16:1, 32:1, 64:1, . . . etc. Accordingly, a determination is made whether the compression is complete (step 16). If the compression is incomplete, further look-up is performed. If the compression is complete, however, the compressed data is transmitted (step 18). It is determined at a gateway whether further compression is required (step 19). If so, transcoding is performed (step 20). The receiver receives the compressed data (step 22).

Abstract: A technique for automatically producing, or training, a set of bitmapped character templates defined according to the sidebearing model of character image positioning uses as input a text line image of unsegmented characters, called glyphs, as the source of training samples. The training process also uses a transcription associated with the text line image, and an explicit, grammar-based text line image source model that describes the structural and functional features of a set of possible text line images that may be used as the source of training samples. The transcription may be a literal transcription of the line image, or it may be nonliteral, for example containing logical structure tags for document formatting and layout, such as found in markup languages.

Abstract: An image decoding and recognition system and method comprising a fast heuristic algorithm using hidden Markov models (HMM). The new search algorithm, called an "iterative complete path" (ICP) algorithm, patterned after well-known branch-and-bound (B&B) methods, significantly reduces the complexity and improves the speed of HMM image decoding without sacrificing the optimality of the straightforward procedure. An advantageous form of the heuristic functions which is useful in applying the ICP algorithm to text-like images is described. The ICP algorithm is directly applicable to the separable type of finite-state source models. Also disclosed is a technique for transforming more general source models into such a separable form.

Abstract: An image recognition system, in particular for document image recognition, using an imaging model employing a 2-dimensional finite state automaton corresponding to a regular string grammar. This approach is not only less computationally intensive than previous grammar-based approaches to document image recognition, but also can handle a wider variety of image types. Features of the imaging model include a sidebearing model of glyph positioning, an image decoder based on linear scheduling theory for regular interative algorithms, the combining of overlapping image sub-regions, and a least-squares estimation procedure for measuring character parameters from character samples in the image.

Abstract: A method of automatically identifying bitmapped image objects. Each of a set of templates in an object template library is compared with all areas of like size of a bitmapped image. A set of signals is generated for each such comparison that satisfies a defined matching criteria between the template and the image area being compared. The set of signals identifies the object based on the matching template, the location of the object in the image and an indication of the goodness of the match between the object and the template. A series of possible parse trees are formed that describe the image with a probability of occurrence for each tree. Each parent node and its child nodes of each parse tree satisfies a grammatical production rule in which some of the production rules define spatial relationships between objects in the image. The one of the possible parse trees which has the largest probability of occurence is selected for further utilization.