Scanner ML Segmentation for Mixed-Content Document Output
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing scanners require user input for selecting image type, leading to suboptimal results when documents contain mixed content, and automatic selection often fails to produce suitable outputs.
Innovation Solution
A scanner equipped with a trained machine learning model that segments and processes document images into regions of different types, applying suitable image processing to each region, with user confirmation and adjustment options.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the scanner requires user selection of image type, then the user can control the processing parameters, but the selection result depends on user perception and may not be appropriate for mixed content documents
Solution Approach 1:
The patent segments the document image into multiple regions with different image types (text regions, photo regions, etc.) and applies different processing parameters to each region. This allows the system to handle mixed content documents by treating each region according to its specific type, resolving the contradiction between user control and processing accuracy.
Solution Approach 2:
The patent implements local quality by applying different image processing parameters to different regions of the document based on their specific types. Text regions receive processing optimized for text clarity, while photo regions receive processing optimized for image quality, thereby achieving high processing accuracy for mixed content without requiring user intervention.
2Ease of operation
If the scanner automatically selects image type, then user operation is simplified, but suitable output results are not always produced for mixed content documents
Solution Approach 1:
The system automatically segments the document into multiple regions and identifies the image type of each region. This segmentation enables the scanner to handle mixed content documents by applying appropriate processing to each region type, achieving both automatic operation and high output quality.
Solution Approach 2:
The system applies different processing parameters to different regions based on their identified image types. Text regions are processed with parameters optimized for text, while photo regions use parameters optimized for images, thereby achieving high output quality for mixed content documents through automatic selection.
3Device complexity
If a single image type is selected for the entire document, then processing is simplified, but documents with mixed content (text and photos) do not produce suitable output results
Solution Approach 1:
The patent divides the document into multiple regions and identifies the image type of each region. This segmentation allows the system to apply different processing parameters to text regions and photo regions separately, achieving suitable output for mixed content documents while maintaining processing simplicity through automated region-based classification.
Solution Approach 2:
The system applies different processing parameters to different regions based on their image types. Text regions receive text-optimized processing while photo regions receive photo-optimized processing, thereby achieving high suitability for mixed content documents without significantly increasing processing complexity.
Data Source
AI summary
A controller of a scanner performs a first scan process in a case where a first scan instruction is received. The first scan process includes scanning a document using a scanning engine to generate first scan data representing a first document image. The controller sends the first scan data to a server through the communication interface. The controller performs a first outputting process in a case where first processed scan data is received from the server. The first processed scan data is generated by a trained machine learning model by processing on one or more target sub-images in accordance with an image type. Each target sub-images is included in a corresponding region in the first document image. The first outputting process includes outputting a first target object. The first target object is the first processed scan data or an object based on the first processed scan data.