Image Processing Apparatus Noise Removal Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods fail to accurately remove scanning noise from images with uniform color areas, leading to incomplete color reduction and degraded editability during vectorization processing.
Innovation Solution
An image processing apparatus that classifies pixels into clusters, specifies background clusters, combines micro-sized label areas with adjacent areas based on characteristic quantities, and generates vector data to effectively remove scanning noise without breaking lines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If only cluster size is used as the criterion for combining clusters, then the processing is simple, but scanning noise cannot be accurately removed and lines may break
Solution Approach 1:
The patent changes the criterion for cluster combination from only size-based to characteristic quantity-based. Specifically, it uses the ratio of the number of pixels on the boundary between different clusters to the total number of pixels in the cluster as a key characteristic quantity. This parameter change enables accurate identification of scanning noise clusters while preserving legitimate color transitions, thereby resolving the contradiction between processing simplicity and noise removal accuracy.
2Manufacturing precision
If a predetermined threshold value is used for cluster size, then small noise clusters are removed, but larger noise clusters remain uncombined
Solution Approach 1:
The patent introduces a new characteristic quantity (boundary pixel ratio) to supplement the traditional size-based threshold approach. By calculating the ratio of boundary pixels to total pixels for each cluster, the system can identify scanning noise clusters regardless of their size. This allows both small and large noise clusters to be detected and combined appropriately, resolving the issue where larger noise clusters were missed by size-based thresholds.
Solution Approach 2:
The patent implements a feedback mechanism where clusters are processed iteratively. After combining clusters in one pass, the algorithm recalculates characteristic quantities and repeats the combination process until no more clusters meet the combination criteria. This feedback loop ensures that even large noise clusters are eventually identified and combined, improving overall noise removal accuracy.
3Loss of information
If color clustering is performed without considering spatial relationships, then color separation is achieved, but scanning noise cannot be distinguished from actual image content
Solution Approach 1:
The patent segments the image processing into distinct stages: color-based clustering, spatial relationship analysis, and characteristic quantity calculation. By separating these functions, the system preserves color information while adding spatial context analysis. The segmentation allows the algorithm to first group pixels by color and then evaluate their spatial distribution to identify scanning noise patterns.
Solution Approach 2:
The patent adds spatial parameters (boundary pixel positions and relationships) to the color-based clustering results. By calculating the ratio of boundary pixels to total pixels and analyzing the spatial distribution of clusters, the system transforms pure color separation into a combined color-spatial analysis, enabling accurate distinction between scanning noise and legitimate image content.
Data Source
AI summary
The cluster of a background is specified among a plurality of clusters classified by clustering processing, and a label area whose size is less than a first threshold value is combined with adjoining another label area which is not the cluster of the background among label areas to which the same label number is given by labeling processing. Then, whether the label area is combined with the adjoining another label area based on characteristic quantity obtained from the label area is determined, and combining, vector data is generated based on the label area after combining the label area when determined as the label area being combined.