Feature Extraction Device Co-occurrence Histogram Discrimination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing feature extraction methods, such as co-occurrence histograms of oriented gradients, face limitations in discrimination power when dealing with multiple target objects or foreground and background in images, due to uniform handling of pixels.
Innovation Solution
A feature extraction device that calculates first and second pixel features, classifies pairs of pixels, and computes co-occurrence frequency using a co-occurrence frequency calculation unit, generating co-occurrence histograms to enhance feature discrimination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixels are handled uniformly in co-occurrence histogram calculation, then the method is simple to implement, but the discrimination power of features is insufficient when multiple target objects or foreground and background are present
Solution Approach 1:
The patent segments pixels into different groups based on their spatial relationships and characteristics. Specifically, it divides pixels into foreground pixels and background pixels, and further segments them into different spatial regions (e.g., left, right, upper, lower regions relative to a reference pixel). This segmentation allows the system to handle different pixel groups with different processing rules, thereby improving discrimination power while maintaining manageable complexity through structured organization.
Solution Approach 2:
The patent applies local quality by assigning different processing characteristics to different spatial regions and pixel types. For example, it calculates co-occurrence histograms differently for pixels in different directions (left, right, upper, lower) and for different pixel types (foreground vs. background). This localized processing approach enables the system to capture region-specific features that improve discrimination power without requiring complete redesign of the entire processing pipeline.
2Loss of information
If co-occurrence histograms are calculated for all pixel pairs, then comprehensive feature extraction is achieved, but computational complexity increases significantly
Solution Approach 1:
The patent reduces computational complexity by segmenting the pixel pairs that need to be processed. Instead of calculating co-occurrence histograms for all possible pixel pairs, it focuses calculations on specific segments: pairs involving a reference pixel and pixels in specific directional regions (left, right, upper, lower), and pairs within foreground and background regions separately. This segmentation maintains comprehensive feature extraction for relevant regions while avoiding redundant calculations.
Solution Approach 2:
The patent applies partial action by selectively calculating co-occurrence histograms only for pixel pairs that are most relevant to object detection and classification. Rather than exhaustively processing all pixel pairs in the image, it focuses computational resources on pairs that are more likely to contribute to discrimination power, such as pairs within foreground regions and pairs in specific directional relationships to reference pixels.
Data Source
AI summary
According to one embodiment, a feature extraction device includes an obtaining unit that obtains image data having a plurality of pixels. The device includes a pixel feature calculation unit that calculates first pixel features and second pixel features of each of the pixels, which are different from each other, and a classification unit that classifies a pair of pixels by using the first features for at least some of the plurality of pixels. The device includes a co-occurrence frequency calculation unit that calculates a co-occurrence frequency representing a frequency of co-occurrence of the second pixel features of the first pixel and the second pixel features of the second pixel for the set for which a result of the classification by the classification unit is consistent.