Michael J Perrone

US Patent:

6567548, May 20, 2003

Filed:

Jan 29, 1999

Appl. No.:

09/240362

Inventors:

Krishna S. Nathan - New York NY
Michael P. Perrone - Yorktown NY
John F. Pitrelli - Danbury CT

Assignee:

International Business Machines Corporation - Armonk NY

International Classification:

G06K 918

US Classification:

382186, 382159, 382179, 382187

Abstract:

A handwriting recognition system and method whereby various character sequences (which are typically âslurredâ together when handwritten) are each modelled as a single character (âcompound character modelâ) so as to provide increased decoding accuracy for slurred handwritten character sequences. In one aspect of the present invention, a method for generating a handwriting recognition system having compound character models comprises the steps of: providing an initial handwriting recognition system having individual character models; collecting and labelling a set of handwriting data; aligning the labelled set of handwriting data; generating compound character data using the aligned handwriting data; and retraining the initial recognition system with the compound character data to generate a new recognition system having compound character models. Once these compound character models are trained, they may be used to accurately decode slurred handwritten character sequences for which compound character models were previously generated. Once recognized, the compound characters are expanded into the constituent individual characters comprising the compound character.

US Patent:

7260779, Aug 21, 2007

Filed:

Sep 30, 2005

Appl. No.:

11/240605

Inventors:

Paul Turquand Keyser - Mount Kisco NY, US
Michael Peter Perrone - Yorktown NY, US
Eugene H. Ratzlaff - Hopewell Junction NY, US
Jayashree Subrahmonia - White Plains NY, US

Assignee:

International Business Machines Corporation - Armonk NY

International Classification:

G06F 15/00

US Classification:

715525, 382187

Abstract:

In one aspect of the present invention, page breaks are identified in the following manner. A set of ink data and a document description are processed by a variety of scoring methods, each of which generates a score for each possible insertion point in the ink. These scores are combined to produce a ranked list of hypothesized page breaks for the corresponding ink data. This ranked list is then used either to insert page breaks automatically using a predefined threshold to determine a cut-off in the list; or to present, on-line, to a human for verification/approval; or a mixture of the two based on two thresholds: one for automatic insertion and the other for human verification. It is to be understood not all scoring methods need be used, that is, one or more of the scoring methods may be used as needed.

US Patent:

7343041, Mar 11, 2008

Filed:

Feb 19, 2002

Appl. No.:

10/079763

Inventors:

Thomas Yu-Kiu Kwok - Washington Township NJ, US
Michael Peter Perrone - Yorktown Heights NY, US

Assignee:

International Business Machines Corporation - Armonk NY

International Classification:

G06K 9/00
G06K 9/62

US Classification:

382187, 382224

Abstract:

A handwritten word is transcribed into a list of possibly correct transcriptions of the handwritten word. The list contains a number of text words, and this list is compared with previously stored set of lists of text words. Based on a metric, one or more nearest neighbor lists are selected from the set. A decision is made, according to a number of combination rules, as to which text word in the nearest neighbor lists or the recently transcribed list is the best transcription of the handwritten word. This best transcription is selected as the appropriate text word transcription of the handwritten word. The selected word is compared to a true transcription of the selected word. Machine learning techniques are used when the selected and true transcriptions differ. The machine learning techniques create or update rules that are used to determine which text word of the nearest neighbor lists or the recently transcribed list is the correct transcription of the handwritten word.

US Patent:

7627596, Dec 1, 2009

Filed:

Feb 19, 2002

Appl. No.:

10/079741

Inventors:

Thomas Yu-Kiu Kwok - Washington Township NJ, US
James Randal Moulic - Poughkeepsie NY, US
Kenneth Blair Ocheltree - Ossining NY, US
Michael Peter Perrone - Yorktown Heights NY, US
John Ferdinand Pitrelli - Danbury CT, US
Eugene Henry Ratzlaff - Hopewell Junction NY, US
Gregory Fraser Russell - Yorktown Heights NY, US
Jayashree Subrahmonia - White Plains NY, US

Assignee:

International Business Machines Corporation - Armonk NY

International Classification:

G06F 17/00

US Classification:

707102, 715268

Abstract:

The techniques in the present invention allow both text and handwritten queries, and the queries can be single-word or multiword. Generally, each handwritten word in a handwritten document is converted to a document stack of words, where each document stack contains a list of text words and a word score of some type for each text word in the list. The query is also converted to one or more stacks of words. A measure is determined from each query and document stack. Documents that meet search criteria in the query are then selected based on the query and the values of the measures. The present invention also performs multiple recognitions, with multiple recognizers, on a handwritten document to create multiple recognized transcriptions of the document. The multiple transcriptions are used for document retrieval. In another embodiment, a single transcription is created from the multiple transcriptions, and the single transcription is used for document retrieval.

US Patent:

7697760, Apr 13, 2010

Filed:

Jan 11, 2008

Appl. No.:

11/972913

Inventors:

Thomas Yu-Kiu Kwok - Washington Township NJ, US
Michael Peter Perrone - Yorktown Heights NY, US

Assignee:

International Business Machines Corporation - Armonk NY

International Classification:

G06K 9/00
G06K 9/18
G06K 9/72

US Classification:

382186, 382187, 382229

Abstract:

A handwritten word is transcribed into a list of possibly correct transcriptions of the handwritten word. The list contains a number of text words, and this list is compared with previously stored set of lists of text words. Based on a metric, one or more nearest neighbor lists are selected from the set. A decision is made, according to a number of combination rules, as to which text word in the nearest neighbor lists or the recently transcribed list is the best transcription of the handwritten word. This best transcription is selected as the appropriate text word transcription of the handwritten word. The selected word is compared to a true transcription of the selected word Machine learning techniques are used when the selected and true transcriptions differ. The machine learning techniques create or update rules that are used to determine which text word of the nearest neighbor lists or the recently transcribed list is the correct transcription of the handwritten word.

US Patent:

8396256, Mar 12, 2013

Filed:

Mar 25, 2010

Appl. No.:

12/731579

Inventors:

Ligang Lu - New City NY, US
Brent Paulovicks - Danbury CT, US
Michael Peter Perrone - Yorktown Heights NY, US
Vadim Sheinin - Mount Kisco NY, US

Assignee:

International Business Machines Corporation - Armonk NY

International Classification:

G06K 9/00
G06F 15/80
G01C 1/00

US Classification:

382113, 345505, 356145

Abstract:

Techniques are disclosed for parallel computing of a line of sight (LoS) map (e. g. , view-shed) in a parallel computing system. For example, a method for computing an LoS map comprises the following steps. Data representing at least one image is obtained. An observation point in the at least one image is identified. A portion of the data that is associated with a given area in the image is partitioned into a plurality of sub-areas. The plurality of sub-areas are assigned to a plurality of processor elements of a parallel computing system, respectively, such that the data associated with each one of the plurality of sub-areas is processed independent from the data associated with each other of the plurality of sub-areas, wherein results of the processing by the processor elements represents the LoS map. The parallel computing system may be a multicore processor.

US Patent:

20120316785, Dec 13, 2012

Filed:

Feb 1, 2012

Appl. No.:

13/363999

Inventors:

Ligang Lu - New City NY, US
Michael P. Perrone - Yorktown Heights NY, US

Assignee:

INTERNATIONAL BUSINESS MACHINES CORPORATION - Armonk NY

International Classification:

G06F 19/00

US Classification:

702 2

Abstract:

A system, method and computer program product for seismic imaging implements a seismic imaging algorithm utilizing Reverse Time Migration technique requiring large communication bandwidth and low latency to convert a parallel problem into one solved using massive domain partitioning. Several aspects of the imaging problem are addressed, including very regular and local communication patterns, balanced compute and communication requirements, scratch data handling and multiple-pass approaches. The partitioning of the velocity model into processing blocks allows each sub-problem to fit in a local cache, increasing locality and bandwidth and reducing latency. The RTM seismic data processing utilizes data that includes combined shot data, i.e., shot data selected from amongst a plurality of shots that are combined at like spatial points of the volume.

US Patent:

20120316786, Dec 13, 2012

Filed:

Feb 1, 2012

Appl. No.:

13/364054

Inventors:

Ligang Lu - New City NY, US
Michael P. Perrone - Yorktown Heights NY, US

Assignee:

INTERNATIONAL BUSINESS MACHINES CORPORATION - Armonk NY

International Classification:

G06F 19/00

US Classification:

702 2

Abstract:

A system and method implementing a hierarchical approach to RTM (Reverse Time Migration) seismic imaging at different granularity in space and time. An RTM seismic imaging algorithm utilizes RTM technique to convert a parallel problem into one solved using massive domain partitioning. In the method, a coarse-grain grid for the 3D volume of the geological subsurface structure under investigation is initially processed, permitting the RTM imaging process to be performed faster and produces lower level seismic image for inspection. Criteria are then applied to the first level of seismic image to determine whether to reject the image or whether a finer resolution seismic imaging is needed. In the case of finer resolution is needed, RTM resolution for the target volume is adjusted accordingly and RTM imaging process is applied with the new resolution. The process is repeated until either the image is accepted or rejected.