Automatic Learning Method for Object Appearance Forms in Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face challenges in efficiently processing a large number of positive and negative training examples, which are necessary for learning the multifaceted forms of appearance of objects and backgrounds, leading to limited robustness and increased computational expense.
Innovation Solution
An automatic learning method that involves generating feature images from training images, determining classification responses, and using weighted summation through linear filtering operations to efficiently combine feature contributions from annotation and classification images, allowing for the extraction of object features without explicit generation of feature data vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large number of positive and negative training examples are processed to learn multifaceted forms of appearance, then the robustness of object detection is improved, but the computational expense and processing time increase
Solution Approach 1:
The patent segments the training image into multiple feature images by dividing it into a plurality of image areas and extracting features from each area. This segmentation allows the system to process training examples in a distributed manner, improving robustness while managing computational complexity through structured processing of smaller feature units.
Solution Approach 2:
The patent transforms the training process from a single-image domain to a multi-dimensional feature space by generating multiple feature images from different image areas. This dimensional transformation enables the system to capture multifaceted forms of appearance more effectively, improving robustness without proportionally increasing computational cost.
2Measurement precision
If feature data vectors are explicitly generated from training images, then object features can be extracted, but the computational complexity and processing time increase
Solution Approach 1:
The patent extracts features directly from image areas to generate feature images, eliminating the need for explicit feature data vector generation. This extraction approach maintains measurement precision by working directly with image data while reducing computational complexity by avoiding intermediate vector representations.
Solution Approach 2:
The patent introduces feature images as an intermediary representation between raw training images and final object features. This intermediary allows for efficient feature extraction and combination without requiring explicit feature data vectors, thereby reducing computational complexity while maintaining accuracy.
3Adaptability or versatility
If training images with multiple overlapping training examples are processed, then the variance of forms of appearance is improved, but the processing time and computational resources increase
Solution Approach 1:
The patent segments overlapping training examples into distinct image areas and processes them in parallel through feature extraction. This segmentation enables the system to handle multiple training examples simultaneously, increasing the variance of learned forms of appearance while managing processing time through efficient parallel processing of segmented features.
Solution Approach 2:
The patent maintains continuous processing of training examples by generating feature images from all image areas without interruption. This continuous action ensures that all training examples contribute to the learning process, maximizing the variance of forms of appearance while minimizing idle processing time through sustained efficient computation.
Data Source
AI summary
An automatic learning method for the automatic learning of the forms of appearance of objects in images in the form of object features from training images for using the learned object features in an image processing system comprises determining a feature contribution by a training image to object features by weighted summation of training image features by means of linear filter operations, applied to the feature image, by using a weight image obtained at least from an annotation image and a classification image. This allows faster learning processes and also the learning of a greater variance of forms of appearance of objects and backgrounds, which increases the robustness of the system in its application to untrained images.