Thomas G Tsao

20040175057, Sep 9, 2004

Mar 4, 2003

10/377649

Thomas Tsao - Potomac MD, US
Stanley Yuen - Montgomery Village MD, US

G06K009/36
G06F015/00

382/294000, 712/010000

An affine transformation analysis system and method is provided for matching two images. The novel systolic array image affine transformation analysis system comprising a linear rf-processing means, an affine parameter incremental updating means, and a least square error fitting means is based on a Lie transformation group model of cortical visual motion and stereo processing. Image data is provided to a plurality of component linear rf-processing means each comprising a Gabor receptive field, a dynamical Gabor receptive field, and six Lie germs. The Gabor coefficients of images and affine Lie derivatives are extracted from responses of linear receptive fields, respectively. The differences and affine Lie-derivatives of these Gabor coefficients obtained from each parallel pipelined linear rf-processing components are then input to a least square error fitting means, a systolic array comprising a QR decomposition means and a backward substitution means. The output signal from the least square error fitting means then be used to updating of the affine parameters until the difference signals between the Gabor coefficients from the static and dynamical Gabor receptive fields are substantially reduced.

20120008868, Jan 12, 2012

Jul 8, 2010

12/832144

Thomas Tsao - Potomac MD, US
Xuemei Cheng - Rockville MD, US

COMPUSENSOR TECHNOLOGY CORP. - Germantown MD

G06K 9/62

382224

A video image event attention and analysis system comprising a video image input means, an event attention means, an event report means and a method comprising numerous steps of processes is provided for automatically detecting, analyzing, classifying events from input video image sequences. The changes in the event happen at some small part of visual world embedded in a largely persistent environment. Visual events are characterized through such duality. Accordingly, the method of this invention comprises processes first extracting features of the persistent environment and features of active agents, both are event specific, in separated channels, each followed by a corresponding recognition process, one for environment and one for active agents, and then the dynamical system of their coupling, particular their movements against environment and against each other, are analyzed and results are synthesized and interpreted in a process of hierarchical event classification.

59110350, Jun 8, 1999

Apr 12, 1995

8/420762

Thomas Tsao - Silver Spring MD

G06F 1518

395 21

A method and apparatus is provided to determine affine disparity and affine invariant pattern distance from two image patterns. The method formulates the problem of determining affine disparity and affine invariant distances between two image patterns as one of finding an affine transform for an adjustable hypercolumn (HC) reference frame that maximally compensates the affine difference of the two image patterns. The image HC-coding and the difference vector is computed from two image patterns using specific orientation (SO) simple cell receptive fields employed in an HC-reference system. The Lie derivative of the difference vector is computed from image patterns using Lie germs. In accordance with one aspect of the invention, the affine transform that maximally compensates the affine difference between two image patterns is found by a gradient system wherein the gradient is calculated from the difference vector and its Lie derivatives. A closed loop adaptive HC-reference frame network is provided to extract binocular affine disparity using an energy minimization technique in a manner which permits for coincidental binocular fusion of the images in the HC-coding and determination of affine invariant pattern distance in general.

20150131897, May 14, 2015

Nov 13, 2013

14/078751

Thomas Tsao - Germantown MD, US
Xuemei Cheng - Rockville MD, US

G06T 7/00
G06K 9/00

382154

A method and apparatus for extracting surface representation from images and video data for segmenting image plane according to the surface connectivity, and identifying areas of images taken by a moving camera according to the object surfaces wherefrom the areas of images are taken, are disclosed. The invention discloses a method and apparatus comprising a plurality of processing modules for extracting from images in a video sequence the occluding contours delineating images into regions in accordance with the spatial connectivity of the correspondent visible surfaces, and diffeomorphism relations between areas of images taken from different perspective centers for identifying image areas of different frames as of the surface of same object, and specifying the fold contours of the surfaces that owns the contour, and thus producing the surface representations from video images taken from persistent objects by a moving camera.