The University of Manchester

Toyota Motor Europe

Imorphics Limited

Timothy Francis Cootes

Christopher John Taylor

Gareth Edwards

Rhodri Davies

Claudia Lindner

Carole Twining

Mircea Ionita

David Cristinacce

Nicholas P Costen

Carole c/o Stopford Building Twining

One embodiment of the invention provides an image processing method for use in fitting a deformable shape model to an acquired image. The deformable shape model specifies a set of target points whose motion is governed by the model. The method comprises, for each target point, generating a corresponding response image by: providing a feature detector to locate a given target point within the acquired image, wherein said feature detector includes a random forest comprising one or more decision trees; scanning a patch image across the acquired image to define multiple sampling locations with respect to the acquired image; for each sampling location of the patch image with respect to the acquired image, performing regression voting using the random forest to produce one or more votes for the location of the given target point within the acquired image; and accumulating the regression votes for all of the multiple sampling locations to generate said response image corresponding to the given target point. The method further includes then performing an optimization to determine positions for the set of target points within the acquired image based on the corresponding response images and subject to constraint by the deformable shape model.

Image processing apparatus and method for fitting a deformable shape model to an image using random forest regression voting

A method of locating features of an object, of a class of objects, of a class of objects, within a target image. The method comprises initializing a set of feature points within the target image, each feature point corresponding to a predetermined feature for objects of the class of objects; deriving a set of template detectors, from the set of feature points, using a statistical model of the class of objects, each template detector comprising an area of image located about the location of a feature point for an object of the class of objects; comparing the set of template detectors with the target image; and updating the set of feature points within the target image in response to the result of the comparison.

Method of locating features of an object

A statistical shape model is built by automatically establishing correspondence between a set of two dimensional shapes or three dimensional shapes. A parameterization of each shape is determined and, a statistical shape model is built using the parameterization. An objective function is used to provide an output which indicates the quality of the statistical shape model. By performing these steps repeatedly for different parameterizations and comparing the quality of the resulting statistical shape models using output of the objective function to determine which parameterization provides the statistical shape model having the best quality, the output of the objective function is a measure of the quantity of information required to code the set of shapes using the statistical shape model.

Constructing a statistical shape model from two-dimensional or three-dimensional data

A method of object identification comprising, for an image in which an object is to be identified, determining for each of a set of location in the direction in which the gradient of intensity change to an adjacent location is greatest together with the magnitude of that gradient, modifying the gradient magnitude using a nonlinear function, the modified gradient and the associated direction for the set of location providing a vector representative of the image, and comparing the vector with a previously generated statistical model which provides identification of the object.

Object identification

A method of identifying an object class uses a model based upon appearance parameters derived by comparing images of objects of different classes. The model includes a representation of a probability density function describing a range over which appearance parameters may vary for a given class of object, the model further including a defined relationship between the appearance parameters and the probability density function. The method generates appearance parameters representative of an unknown object, estimates an appropriate probability density function for the unknown object using the defined relationship between the appearance parameters and the probability density function, then iteratively modifies at least some of the appearance parameters within limits determined using the probability density function to identify the object class.

Object class identification, verification of object image synthesis

A method of determining face sub-spaces includes making initial estimates of the sub-spaces, for example lighting, pose, identity and expression, using Principle Component Analysis on appropriate groups of faces. An iterative algorithm is applied to image codings to maximize the probability of coding across these non-orthogonal sub-spaces, obtaining the projection on each sub-space, and recalculating the spaces.

Growing community of inventors

Stockport, United Kingdom

Timothy Francis Cootes