Video Processing Apparatus Pixel Classification Error Rate Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The extraction accuracy of subject regions in videos is not 100%, leading to erroneous extraction or holes in the subject image due to misclassification of pixels, which deteriorates the subjective quality of the extracted image.
Innovation Solution
A video processing device and method that classify each pixel as foreground, background, or unclassifiable, calculate an error rate for unclassifiable pixels, and output a subject image with an evaluated value representing the difficulty of classification, while superimposing effects on pixels with higher evaluated values to improve subjective quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If subject extraction is performed using background difference or machine learning, then subject regions can be extracted from video frames, but extraction accuracy is less than 100% leading to erroneous extraction or holes in the subject image
Solution Approach 1:
The patent implements feedback by evaluating the reliability of each pixel classification result and using this information to iteratively improve the extraction. The reliability evaluation unit assesses whether pixels are classified with high confidence, and this feedback is used to identify unclassifiable pixels that require further processing or correction, thereby improving both extraction accuracy and subjective quality.
Solution Approach 2:
The patent changes the parameter of pixel classification by introducing reliability evaluation and identifying unclassifiable pixels. Instead of binary classification (foreground/background), the system adds a third state (unclassifiable) based on reliability thresholds, allowing for more nuanced handling of ambiguous pixels and reducing erroneous extraction while maintaining subject integrity.
2Productivity
If foreground labels are assigned to extract subject images, then subject regions can be isolated, but errors in label assignment cause holes or erroneous extraction that deteriorate image quality
Solution Approach 1:
The patent introduces reliability evaluation as an intermediary step between pixel classification and final label assignment. This intermediary mechanism assesses the confidence of each classification decision, allowing the system to identify and handle uncertain cases separately, thereby improving classification precision without significantly impacting extraction efficiency.
Solution Approach 2:
The patent segments the pixel population into three distinct groups: foreground pixels, background pixels, and unclassifiable pixels. This segmentation based on reliability evaluation allows for differentiated processing of each group, improving overall classification precision by focusing additional processing resources on the ambiguous unclassifiable pixels that contribute most to extraction errors.
3Device complexity
If all pixels are classified as foreground or background, then extraction process is simple and fast, but classification errors lead to poor subjective quality
Solution Approach 1:
The patent applies local quality by treating different pixels differently based on their classification reliability. Instead of uniform processing, the system identifies unclassifiable pixels locally and applies specialized handling only to these problematic regions, maintaining simple processing for confident classifications while improving reliability for ambiguous cases through targeted additional processing.
Data Source
AI summary
A video processing device 1 includes: a foreground extraction unit 12 configured to classify each pixel in an input image as foreground, background or unclassifiable; an error rate evaluation unit 13 configured to obtain an error rate for unclassifiable pixels based on previous classification results to calculate an evaluated value representing difficulty of classification; a processing unit 14 configured to arrange an effect to be superimposed on a subject image classified as foreground in accordance with the evaluated value; and the output unit 15 configured to output an output image obtained by superimposing the effect on the subject image.