Object shape alignment device, object processing device and methods thereof

Abstract

The invention relates to an object shape alignment device, an object processing device and methods thereof. The object shape alignment device comprises a unit obtaining an object shape regression model, which contains an average object shape, a plurality of regression functions and a plurality of characteristic selection graphs, from a plurality of training samples, a unit setting an initial object shape on the basis of the average object shape, a unit calculating at least one characteristic vector of a plurality of characteristic points of the initial object shape, a unit selecting, for each coordinate of each characteristic point of the initial object shape, a characteristic fragment from the calculated characteristic vectors on the basis of the corresponding one in the characteristic selection graphs and assembling the characteristic fragments into sub characteristic vectors, and a unit predicting, for at least one coordinate of the at least one characteristic point of the initial object shape, a coordinate increment on the basis of a corresponding one of the corresponding sub characteristic vector and the regression functions. The devices and methods can reduce model dimensions and achieve high precision, speed, robustness and / or the like.

Description

technical field [0001] The present invention generally relates to the fields of image processing, computer vision and pattern recognition, and in particular relates to an object shape alignment device, an object processing device and a method thereof. Background technique [0002] In the fields of image processing, computer vision, and pattern recognition, automatically and precisely aligning object shapes (or detecting feature points) described by a set of feature points is a key task, and this can be widely used for example in facial recognition, Gesture recognition, expression analysis, 3D facial simulation, facial cartoon animation, etc. [0003] Current object shape alignment methods adopt model-based methods (such as Active Shape Model (ASM) and Active Appearance Model (AAM)) or regression-based methods (such as explicit regression (ExplicitShapeRegression, ESR) and supervised descent method (Supervised Descent Method, SDM)). [0004] Since object shape alignment is ...

AI Technical Summary

Problems solved by technology

This high-dimensional feature vector is highly redundant for the alignment process and thus makes the model size or dictionary size too large

Images