Filled Form Field Label Detection with OCR-Tolerant Histograms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques struggle to accurately identify field labels in filled forms due to variations in scale, rotation, font style, and font size, and require substantial data sets for AI training, making it time-consuming and difficult.
Innovation Solution
A string encoding technique and field label-identifying histogram hierarchical method that tolerates OCR errors, allowing for robust alignment and identification of field labels without extensive AI training data, using bounding boxes, text scaling, and histogram analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If image processing techniques are applied to filled forms with variations in scale, rotation, font style, and font size, then field label identification can be performed, but the accuracy and reliability of identification deteriorates due to significant differences between form images
Solution Approach 1:
The patent applies parameter changes by normalizing image parameters (scale, rotation, font size, font style) through transformations before processing. The system adjusts images to a standard parameter set, enabling reliable field label identification despite original variations in these parameters across different filled forms.
Solution Approach 2:
The patent performs preliminary actions by pre-processing and normalizing form images before field label identification. This includes adjusting scale, rotation, and font parameters in advance, so that the subsequent identification process can operate on standardized images, improving reliability.
2Extent of automation
If AI techniques are used to identify field labels in filled forms, then identification can be achieved, but the process becomes time consuming and difficult to accomplish requiring substantial training data
Solution Approach 1:
The patent replaces complex AI-based mechanical systems with simpler image processing techniques. Instead of using trained neural networks or machine learning models that require substantial data and time, the system uses direct image processing methods including feature extraction, template matching, and parameter normalization to achieve field label identification.
Solution Approach 2:
The patent uses copying by creating standardized templates of form fields and matching them against processed images. Rather than training AI models on diverse data, the system copies standardized field patterns and uses them to identify corresponding fields in filled forms, significantly reducing training requirements.
3Productivity
If standard image processing techniques are applied directly to filled forms, then processing can be performed, but the techniques fail to account for variations in scale, rotation, and font characteristics
Solution Approach 1:
The patent applies dynamics by making the image processing system adaptive to variations in scale, rotation, and font characteristics. The system dynamically adjusts processing parameters based on detected form characteristics, allowing both fast processing and high accuracy by responding to the specific properties of each filled form.
Solution Approach 2:
The patent uses parameter changes to normalize image characteristics before processing. By adjusting scale, rotation, and font parameters to standard values, the system enables standard image processing techniques to work effectively on diverse filled forms, achieving both productivity and precision.
Data Source
AI summary
Method and apparatus to identify field labels from filled forms using image processing to compare two or a few copies of the same kind of form, possibly filled out differently. Filled forms are processed to provide text strings, one for each line of text on each filled form. The text strings are converted to vectors. Vectors from different filled forms are compared to identify common words which are indicative of field labels. In an embodiment, a histogram may be generated to show frequency of occurrence of characters and words, the histogram values also being indicative of field labels.