Address Block and Barcode Detection Using Connected Component Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for detecting and reading address blocks and barcodes on captured images of items, such as mailpieces or parcels, are time-consuming and resource-intensive, particularly when using optical character recognition (OCR), and require significant memory and processor resources.
Innovation Solution
Utilizing machine-learning based connected component analysis to efficiently and accurately detect and read address blocks and barcodes by performing unsupervised machine learning on binarized images, including density-based spatial clustering (DBSCAN) and k-means methods, to cluster objects and identify the feature of interest.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical character recognition (OCR) is used to detect and read address blocks and barcodes, then reading accuracy is improved, but processing time and resource consumption increase significantly
Solution Approach 1:
The patent segments the image processing task into distinct phases: binarization, connected component analysis to identify candidate regions, and selective OCR application. By dividing the full image into discrete components and applying OCR only to identified address block and barcode regions rather than the entire image, processing time is reduced while maintaining reading accuracy.
Solution Approach 2:
The patent extracts only the relevant features (address blocks and barcodes) from the captured image using connected component analysis on binarized images. By taking out and isolating these specific regions of interest, the system avoids applying resource-intensive OCR to the entire image, thereby reducing processing time while preserving reading accuracy on the extracted features.
2Measurement precision
If optical character recognition (OCR) is used to detect and read address blocks and barcodes, then reading accuracy is improved, but memory and processor resources increase significantly
Solution Approach 1:
The patent segments the image processing task into distinct phases: binarization, connected component analysis to identify candidate regions, and selective OCR application. By dividing the full image into discrete components and applying OCR only to identified address block and barcode regions rather than the entire image, processing time is reduced while maintaining reading accuracy.
Solution Approach 2:
The patent extracts only the relevant features (address blocks and barcodes) from the captured image using connected component analysis on binarized images. By taking out and isolating these specific regions of interest, the system avoids applying resource-intensive OCR to the entire image, thereby reducing processing time while preserving reading accuracy on the extracted features.
3Device complexity
If traditional image processing methods are used to detect address blocks and barcodes, then system complexity is reduced, but detection accuracy and efficiency decrease
Solution Approach 1:
The patent introduces binarization and connected component analysis as intermediary steps between image capture and OCR. These intermediary processing stages convert the captured image into a simplified binary representation, identify candidate regions through component analysis, and then guide selective OCR application. This intermediary approach maintains relatively simple system architecture while significantly improving detection efficiency and accuracy.
Data Source
AI summary
The present disclosure relates to a system and method for detecting an address block and barcode on a captured image of an item, and reading the detected barcode using connected component analysis. In one aspect, the method includes binarizing a captured image to generate a binarized image having pixel values, inverting the pixel values of the binarized image, processing the inverted pixel values, and filtering the processed image by area. The method may also include machine learning the processed image to cluster objects in the processed image into a plurality of groups using the filtered image, and determining a number of objects in each of the clustered groups by performing additional operations on the clustered objects to detect the address block. The method may also include selecting the group having the greatest number of objects as the address block, and extracting the selected address block.