Image Segmentation Using Cost-Optimized Function and Connected Component Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional segmentation methods in mixed raster content (MRC) image coding and decoding systems often result in errors, such as text being misclassified as background and background being misclassified as foreground, leading to distortion and inefficiencies in image compression.
Innovation Solution
The use of cost-optimized segmentation (COS) and connected component classification (CCC) to divide images into blocks, accurately classify pixels into foreground and background layers, and modify segmentation candidates based on a cost function that prioritizes segmentation edges, spatial smoothness, and accurate content representation, thereby reducing errors and improving compression efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional segmentation methods are used in MRC encoding, then the encoding process can proceed, but segmentation errors occur such as text being misclassified as background and background being misclassified as foreground
Solution Approach 1:
The patent implements feedback mechanisms where the segmentation results are evaluated against ground truth data, and the segmentation algorithm is iteratively refined based on performance metrics. This allows the system to learn from errors and improve segmentation accuracy over time, reducing misclassification of text and background regions.
Solution Approach 2:
The patent adjusts segmentation parameters such as threshold values, block sizes, and classification criteria to optimize the distinction between text and background regions. By dynamically changing these parameters based on image characteristics, the system achieves more reliable segmentation while minimizing information loss.
2Productivity
If segmentation is performed to create binary mask layer, foreground layer, and background layer, then image compression efficiency improves, but segmentation errors cause distortion in the decoded image
Solution Approach 1:
The patent performs preliminary actions by pre-processing the image to enhance text and background region characteristics before segmentation. This includes steps such as noise reduction, contrast enhancement, and edge detection that prepare the image for more accurate segmentation, thereby reducing distortion in the final decoded image while maintaining compression efficiency.
Solution Approach 2:
The patent introduces an intermediary verification step where segmentation candidates are evaluated and refined before final classification. This intermediary process acts as a mediator between the segmentation algorithm and the final binary mask creation, allowing errors to be corrected before they propagate to the decoded image, thus preserving quality while maintaining compression benefits.
3Measurement precision
If the binary mask layer is used to represent fine detail of text fonts, then text encoding accuracy improves, but any segmentation errors cause distortion in the decoded image
Solution Approach 1:
The patent applies partial or excessive action by performing multiple passes of segmentation and validation, and by using more stringent classification criteria than minimally required. This ensures that text edges are captured with high precision while filtering out false positives, thereby maintaining both accuracy and consistency in the binary mask layer.
Solution Approach 2:
The patent replaces simple mechanical thresholding with more sophisticated classification mechanisms that consider multiple features such as edge orientation, texture, and contextual information. This substitution of the segmentation mechanism allows for more reliable and consistent text region identification, reducing distortion while maintaining fine detail accuracy.
Data Source
AI summary
An apparatus usable in an image encoding and decoding system includes segmentation unit to divide an image into one or more blocks, to segment the blocks into a binary mask layer of a foreground and a background according to a cost optimized function and a feature vector to generate a segmentation image according to the segmented blocks.