Document Boundary Detection in Mobile Image Capture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image capture techniques on mobile devices are inefficient due to the inclusion of background elements in images, which increase file size and complicate image processing operations like perspective transform and optical character recognition (OCR), and fail to accurately detect document boundaries, especially when the camera is not perpendicular to the document.
Innovation Solution
A method involving down-sampling, median filtering, Canny edge detection, and Hough transform to identify and group similar line segments, followed by computing a representative quadrilateral for the document borders, which improves document boundary detection and removal of non-document regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional image capture processes are used to capture documents, then the document can be captured in the image, but the background is also included which increases file size and complicates image processing
Solution Approach 1:
The patent extracts and removes non-document regions from the captured image by detecting document boundaries and creating a mask that isolates only the document area. This extraction process eliminates background elements while preserving the document, thereby reducing file size without sacrificing document capture accuracy
Solution Approach 2:
The image processing pipeline segments the captured image into document regions and non-document regions through multiple processing stages including down-sampling, filtering, edge detection, and line segment analysis. This segmentation enables selective retention of document areas and removal of background, resolving the contradiction between complete document capture and file size reduction
2Reliability
If traditional image capture processes are used to capture documents, then the document can be captured in the image, but the background causes undesirable problems with subsequent image processing operations such as perspective transform and OCR
Solution Approach 1:
By extracting and removing non-document regions through boundary detection and masking, the patent eliminates background interference that complicates perspective transform and OCR operations. The extracted document region is then processed independently, simplifying subsequent operations while maintaining document capture accuracy
Solution Approach 2:
The patent performs preliminary document boundary detection and region extraction before executing perspective transform and OCR operations. This preliminary action prepares the image by isolating the document area and removing background, thereby simplifying subsequent processing steps and improving their effectiveness
3Measurement precision
If the camera position and zoom are adjusted to capture the document without background, then the document boundaries can be captured accurately, but the process is tedious and inefficient
Solution Approach 1:
The patent implements automatic document boundary detection that performs the task of manual camera positioning and framing. The system autonomously analyzes the captured image, detects document boundaries through image processing algorithms, and identifies the document region without requiring user intervention, thereby achieving precise boundary detection while eliminating the time-consuming manual adjustment process
Solution Approach 2:
The patent replaces the mechanical process of manual camera positioning and zoom adjustment with an automated image processing system. Instead of physically adjusting the camera to frame the document perfectly, the system captures the document along with background and then uses algorithmic boundary detection to identify and extract the document region, substituting mechanical precision requirements with computational analysis
Data Source
AI summary
A computer-implemented method can include receiving an input image of a physical document, performing down-sampling on the input image, applying median filtering to the input image, applying Canny edge detection to the input image, performing a Hough transform on the input image, computing a quadrilateral having sides that represent borders of the physical document, and providing the computed quadrilateral as an output.