Object Shape Aligning Apparatus Using L1-Norm Regularization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object shape aligning methods, such as the Supervised Descent Method (SDM), face challenges with high dimensionality of feature vectors leading to multicolinearity, feature redundancy, and over-fitting, making model training unstable and requiring vast training samples.
Innovation Solution
The proposed solution involves using a L1-norm regularized linear regression method with Least Angle Regression and Lasso modification to select feature fragments, reducing the dimensionality of feature vectors and employing specific sub-feature vectors and regression functions for each coordinate, thereby reducing model size and improving accuracy and robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If high dimensional feature vectors are used in SDM, then feature representation richness is improved, but multicolinearity and model training stability deteriorate
Solution Approach 1:
The patent extracts and removes redundant features from the high-dimensional feature vector through L1-norm regularization. The regression model automatically identifies and eliminates multicollinear features, retaining only the most informative subset. This extraction process maintains feature representation richness while removing harmful redundancies that cause training instability.
Solution Approach 2:
The patent changes the parameter representation by introducing L1-norm regularization to the loss function. This parameter change transforms the optimization landscape, enabling the model to selectively zero out redundant feature coefficients while preserving important features. The regularization parameter controls the degree of feature selection, balancing representation richness with training stability.
2Loss of information
If high dimensional feature vectors are used in SDM, then feature completeness is improved, but feature redundancy increases
Solution Approach 1:
The patent extracts essential features from the complete feature set by applying L1-norm regularization. This process identifies and removes redundant features while preserving the complete information necessary for accurate shape alignment. The regularization technique automatically selects the minimal sufficient feature subset.
Solution Approach 2:
The patent discards redundant features through L1-norm regularization that drives their coefficients to zero. The important features are recovered and retained in the final model. This discarding and recovering process eliminates feature redundancy while maintaining feature completeness for accurate alignment.
3Measurement precision
If high dimensional feature vectors are used in SDM, then alignment accuracy is improved, but model size increases
Solution Approach 1:
The patent extracts only the necessary features for accurate alignment by applying L1-norm regularization. This extraction creates a compact model that retains alignment accuracy by keeping only the most informative features. The resulting model is smaller because redundant features have been removed, yet maintains precision through selective feature retention.
Solution Approach 2:
The patent changes the model parameters through L1-norm regularization, which transforms the dense parameter vector into a sparse one. This parameter change reduces model size by setting unnecessary parameters to zero while preserving the parameters that contribute most to alignment accuracy. The regularization strength controls the trade-off between model compactness and accuracy.
4Loss of information
If high dimensional feature vectors are used in SDM, then feature detail richness is improved, but over-fitting increases
Solution Approach 1:
The patent extracts the most generalizable features from the detailed feature set through L1-norm regularization. This extraction process removes features that lead to over-fitting while preserving the detailed information necessary for accurate alignment. The regularization technique promotes sparsity, which acts as a form of regularization that prevents over-fitting by eliminating redundant detailed features.
Solution Approach 2:
The patent changes the parameter distribution through L1-norm regularization, transforming a dense parameter vector into a sparse one. This parameter change reduces over-fitting by zeroing out parameters that capture noise or dataset-specific patterns while retaining parameters that represent generalizable features. The regularization parameter controls the balance between utilizing feature detail and preventing over-fitting.
Data Source
AI summary
An apparatus includes a unit configured to acquire an object shape regression model, which comprises an average object shape, a plurality of regression functions and a plurality of feature selection maps; a unit configured to set an initial object shape for the object image based on the average object shape; a unit configured to calculate at least one feature vector with respect to a plurality of feature points; a unit configured, for each coordinate of the plurality of feature points, to select feature fragments from the calculated feature vector based on a corresponding one of the plurality of feature selection maps and assemble the feature fragments into a sub feature vector; and a unit configured, for at least one coordinate of at least one feature point, to predict a coordinate increment based on the corresponding sub feature vector and a corresponding one of the plurality of regression functions.