Foreground Detection Accuracy via Visual Element Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing foreground detection methods in image processing suffer from false detection due to changes in background elements like water ripples or leaves, and low accuracy in graph segmentation algorithms, which reduces the reliability of distinguishing foreground from background.
Innovation Solution
An image processing apparatus that acquires a current image and a background model, determining similarity measures between visual elements in the current image and the background model to classify them as foreground or background, utilizing neighboring visual elements classified as background in previous images to improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If background subtraction techniques are used to classify visual elements, then foreground detection can be performed, but false foreground detection occurs when background elements change (e.g., water ripples, leaves moving in wind)
Solution Approach 1:
The patent segments the image into visual elements (super-pixels) and processes each element individually. By dividing the image into manageable units and analyzing their temporal consistency separately, the system can distinguish between actual foreground objects and background elements that change over time, thereby reducing false detection while maintaining detection reliability.
Solution Approach 2:
The patent performs preliminary classification of visual elements as foreground or background before final detection. By pre-classifying visual elements based on their temporal consistency across multiple frames and using this classification to adjust similarity thresholds, the system prepares the detection process in advance, preventing false foreground detection from occurring in the first place.
2Ease of manufacture
If graph segmentation algorithm with low accuracy is used to obtain super-pixels, then visual elements can be obtained, but the visual elements cannot keep constant with corresponding visual elements from previous images
Solution Approach 1:
The patent dynamically adjusts the similarity threshold for each visual element based on its classified type (foreground or background). For background elements that should remain consistent, a higher threshold is applied, while for foreground elements that may change, a lower threshold is used. This dynamic adjustment compensates for the instability introduced by low-accuracy segmentation algorithms.
Solution Approach 2:
The patent changes the similarity parameter dynamically based on the classified type of each visual element. By modifying the similarity threshold parameter according to whether an element is classified as foreground or background, the system adapts to variations in visual element consistency caused by imperfect segmentation algorithms.
3Ease of operation
If background confidence is calculated based on neighboring visual elements, then classification can be performed, but when neighboring elements are falsely detected as foreground, the background confidence becomes smaller and less than threshold
Solution Approach 1:
The patent uses feedback from the classified type of each visual element to adjust the similarity threshold for its neighbors. When a visual element is classified as background, this classification feedback is used to lower the similarity threshold for its neighboring elements, making them more likely to be classified as background as well. This feedback mechanism prevents false foreground detection from propagating through the image.
Solution Approach 2:
The patent applies different similarity thresholds to different regions of the image based on the local classification of visual elements. Instead of using a uniform threshold, the system adjusts the threshold locally for each visual element and its neighbors based on their classified types, thereby maintaining background confidence reliability even when some neighboring elements are falsely detected.
Data Source
AI summary
An image processing apparatus including a unit configured to acquire a current image from an inputted video and a background model which comprises a background image and foreground/background classification information of visual elements; a unit configured to determine first similarity measures between visual elements in the current image and the visual elements in the background model; and a unit configured to classify the visual elements in the current image as the foreground or the background according to the current image, the background image in the background model and the first similarity measures. Wherein, the visual elements in the background model are the visual elements whose classification information is the background and which neighbour to corresponding portions of the visual elements in the current image. Accordingly, the accuracy of the foreground detection could be improved.