Soft Decision Function for Video Pattern Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image pattern recognition methods for video data suffer from high false positive and false negative detection errors, particularly in real-time video segments, due to binary decision-making processes that do not account for uncertainties in feature values.
Innovation Solution
The introduction of a soft decision function with scale values and non-binary decision functions for multiple non-leaf nodes in decision trees, allowing for proportional contribution computation to classifier inputs, which reduces detection errors without requiring extensive retraining or additional training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If binary decision functions are used in decision trees for pattern recognition, then the decision-making process is simple and fast, but detection errors (false positives and false negatives) increase due to inability to account for uncertainties in feature values
Solution Approach 1:
The patent transforms binary decision values into continuous probability values using sigmoid functions. Each node in the decision tree outputs a probability between 0 and 1 instead of a binary decision, allowing the system to account for uncertainties in feature values. This parameter change from discrete to continuous values directly improves detection accuracy by providing nuanced decision-making that reflects the confidence level of each classification.
2Reliability
If soft decision functions with scale values are introduced for multiple non-leaf nodes, then detection errors are reduced by accounting for feature value uncertainties, but computation complexity increases
Solution Approach 1:
The patent applies soft decision functions selectively to non-leaf nodes in the decision tree rather than uniformly to all nodes. By focusing the computationally intensive sigmoid calculations on internal decision nodes where uncertainty propagation has the greatest impact, the system achieves improved detection accuracy without unnecessarily increasing computation power for leaf nodes or root nodes where simpler operations suffice.
3Measurement precision
If proportional contribution computation is used for classifier inputs based on soft decision results, then pattern recognition accuracy improves, but the processing time and computational load increase
Solution Approach 1:
The patent maintains continuous probability values throughout the decision tree evaluation process, allowing for smooth propagation of uncertainty information from root to leaf nodes. This continuous approach enables proportional contribution computation where each path's probability directly weights its contribution to the final classification, improving accuracy without requiring discrete intermediate steps that would increase processing time.
Data Source
Figure 1~1a
Figure 2~3
Figure 4
AI summary
An image pattern recognition method detects a pattern in a sequence of video images or individual images from detected interest points. Feature vectors are extracted with video data from video regions around the interest points. A forest of decision trees is used to compute a set of bin values in histograms with bins corresponding to leaf nodes of the decision trees. Each bin value is a sum of contributions computed for individual interest points. Non-binary decision functions are used to compute the contributions and node dependent scale values are used to compute the arguments of the non-binary decision functions. The node dependent scale values may be computed from standard deviations of feature values found for the nodes, multiplied by a factor that is common to the nodes. This factor may be adjusted by feedback so that it can be set differently for different detection classes.